fix(opencode): default-export the Plugin function directly

OpenCode's plugin loader iterates Object.entries(mod) and invokes every
export as a Plugin factory `(input) => Promise<Hooks>`, deduping by
identity. Our prior default export was a PluginModule object
(`{ id, server }`), which opencode tried to call as a function and
crashed with `fn3 is not a function. (In 'fn3(input)', 'fn3' is an
instance of Object)` at load time.

Default-export the HindsightPlugin function itself so both default and
named `HindsightPlugin` exports point to the same reference (dedupe
suppresses a second call). Update the default-export smoke test to
assert this invariant.

Verified end-to-end against opencode 1.1.49 with the built dist — the
plugin now initializes, registers tools/hooks, and processes session
events without error.

.claude-plugin/marketplace.json

@@ -0,0 +1,15 @@
+{
+  "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
+  "name": "hindsight",
+  "description": "Official Hindsight integrations for Claude Code",
+  "owner": {
+    "name": "vectorize-io"
+  },
+  "plugins": [
+    {
+      "name": "hindsight-memory",
+      "description": "Automatic long-term memory for Claude Code via Hindsight",
+      "source": "./hindsight-integrations/claude-code"
+    }
+  ]
+}

.claude/skills/code-review/SKILL.md

@@ -0,0 +1,205 @@
+---
+name: code-review
+description: Review changed code against project standards. Checks for missing tests, dead code, type safety, lint issues, and coding conventions. Run after completing any implementation work.
+user_invocable: true
+---
+
+# Code Review
+
+Review all changed code against the project's quality standards and coding conventions.
+
+## Code Standards
+
+Read and internalize these standards before writing code. The review steps below verify compliance.
+
+### Python Style
+- Python 3.11+, type hints required
+- Async throughout (asyncpg, async FastAPI)
+- Pydantic models for request/response
+- Ruff for linting (line-length 120)
+- No Python files at project root - maintain clean directory structure
+- **Never use multi-item tuple return values** — not even for internal/private functions. Always use a dataclass or Pydantic model. No exceptions, no "it's just two values" shortcuts. If a function returns more than one value, define a named type for it.
+
+### Type Safety with Pydantic Models
+**NEVER use raw `dict` types for structured data** — this applies to all code, including internal helpers and private functions. If the dict has known keys, it must be a dataclass or Pydantic model:
+- Use Pydantic `BaseModel` for all data structures passed between functions
+- Use `@dataclass` for lightweight internal data containers when Pydantic validation isn't needed
+- Add `@field_validator` for type coercion (e.g., ensuring datetimes are timezone-aware)
+- Avoid `dict.get()` patterns - use typed model attributes instead
+- Parse external data (JSON, API responses) into Pydantic models at the boundary
+- This catches type errors at parse time, not deep in business logic
+- The only acceptable `dict` usage is for truly dynamic/unknown keys (e.g., arbitrary metadata, JSON blobs with no fixed schema)
+
+```python
+# BAD - error-prone dict access
+def process(data: dict) -> str:
+    return data.get("name", "")  # No validation, silent failures
+
+# GOOD - typed and validated
+class UserData(BaseModel):
+    name: str
+    created_at: datetime
+
+    @field_validator("created_at", mode="before")
+    @classmethod
+    def ensure_tz_aware(cls, v):
+        if isinstance(v, str):
+            v = datetime.fromisoformat(v.replace("Z", "+00:00"))
+        if v.tzinfo is None:
+            return v.replace(tzinfo=timezone.utc)
+        return v
+
+def process(data: UserData) -> str:
+    return data.name  # Type-safe, validated at construction
+```
+
+### TypeScript Style
+- Next.js App Router for control plane
+- Tailwind CSS with shadcn/ui components
+
+### Code Comments
+- **Always comment non-trivial technical decisions** with the reasoning behind the choice. If someone would ask "why is it done this way?", there should be a comment.
+- **Keep comments up to date with history** — when changing an approach, update the comment to explain what was tried before and why it was changed. Comments serve as a tracker of previous implementations that likely had problems.
+- Don't comment obvious code — only where the "why" isn't self-evident from the code itself.
+
+```python
+# BAD - no context for future readers
+results = await asyncio.gather(*tasks, return_exceptions=True)
+
+# GOOD - explains the non-obvious choice
+# Use return_exceptions=True to avoid cancelling sibling tasks on failure.
+# Previously we used TaskGroup but it cancelled all tasks when one failed,
+# causing partial writes that left orphaned entity links (see #412).
+results = await asyncio.gather(*tasks, return_exceptions=True)
+```
+
+### Branch Hygiene
+- **Always start new feature branches from `origin/main`** — rebase to ensure a clean base.
+- **Only include commits relevant to the PR/branch/feature** — no unrelated changes. If the branch contains commits that don't belong, they must be removed before merging.
+
+### General Principles
+- Don't add features, refactor code, or make "improvements" beyond what was asked
+- Don't add unnecessary error handling for impossible scenarios
+- Don't create helpers or abstractions for one-time operations
+- No backwards-compatibility hacks (unused vars, re-exports, "removed" comments)
+- Three similar lines of code is better than a premature abstraction
+
+## Review Steps
+
+### 1. Check branch hygiene
+
+- Run `git log --oneline main..HEAD` to list all commits on the branch.
+- Verify every commit is relevant to the feature/PR. Flag any unrelated commits.
+- Check the branch is based on a recent `origin/main` (no stale base).
+
+### 2. Identify changed files
+
+Run `git diff --name-only HEAD` (unstaged) and `git diff --cached --name-only` (staged) to get all changed files. If there are no local changes, diff against the base branch using `git diff main...HEAD --name-only` and `git diff main...HEAD` to review all commits on the current branch.
+
+### 3. Run linters
+
+```bash
+./scripts/hooks/lint.sh
+```
+
+Report any failures. Do NOT fix them yourself — just report.
+
+### 4. Check for dead code
+
+For each changed Python file, check for:
+- Unused imports (Ruff should catch these, but verify)
+- Functions/methods/classes that were added but are never called from anywhere
+- Variables assigned but never read
+- Commented-out code blocks that should be removed
+
+For each changed TypeScript file, check for:
+- Unused imports
+- Unused variables or functions
+- Commented-out code
+
+### 5. Check type safety (Python)
+
+For each changed Python file, check for violations:
+- **No raw `dict` for structured data** — must use Pydantic model or dataclass, even for internal/private functions (only exception: truly dynamic/unknown keys)
+- **No multi-item tuple returns** — must use dataclass or Pydantic model, even for internal/private functions (no exceptions)
+- **Missing type hints** on function parameters and return types
+- **Missing `@field_validator`** for datetime fields that should be timezone-aware
+
+### 6. Check for missing tests
+
+For each new or significantly changed function/endpoint/class:
+- Check if there is a corresponding test addition or update
+- New API endpoints MUST have integration tests
+- New utility functions MUST have unit tests
+- Bug fixes SHOULD have a regression test
+
+Flag any new logic that lacks test coverage.
+
+### 7. Check API consistency
+
+If any files in `hindsight-api-slim/hindsight_api/api/` were changed:
+- Were the OpenAPI specs regenerated? (`./scripts/generate-openapi.sh`)
+- Were the client SDKs regenerated? (`./scripts/generate-clients.sh`)
+- Were the control plane proxy routes updated? (`hindsight-control-plane/src/app/api/`)
+
+### 8. Check code comments
+
+For each non-trivial change:
+- **New non-obvious logic** — is there a comment explaining the reasoning?
+- **Changed approach** — does the comment include what was done before and why it changed?
+- **Stale comments** — do existing comments near the changed code still accurately describe the behavior?
+
+### 9. Check integration completeness
+
+If any files in `hindsight-integrations/` were added or changed, verify:
+- **Tests exist** — the integration must have tests that simulate/exercise the external framework (not just pure unit tests of helpers). Check for a `tests/` directory with meaningful test files.
+- **CI job exists** — check `.github/workflows/test.yml` for a corresponding `test-<name>-integration` job. If missing, flag it.
+- **Release process** — check that the integration name is in the `VALID_INTEGRATIONS` array in `scripts/release-integration.sh`. If missing, flag it.
+- **Code standards** — the integration code must follow all Python style rules (type hints, no raw dicts, no tuple returns, etc.).
+
+### 10. Check MCP tool registration completeness
+
+If any new MCP tools were added or existing tools renamed in `hindsight-api-slim/hindsight_api/mcp_tools.py`:
+- **`_ALL_TOOLS` set** in `mcp_tools.py` — must include the new tool name
+- **`tools_to_register` default set** in `register_mcp_tools()` in `mcp_tools.py` — must include the new tool name
+- **`_SINGLE_BANK_TOOLS` set** in `hindsight-api-slim/hindsight_api/api/mcp.py` — must include the new tool if it is bank-scoped (not a bank-management tool like `list_banks`/`create_bank`)
+- **`MCP_TOOL_GROUPS`** in `hindsight-control-plane/src/components/bank-config-view.tsx` — must include the new tool in the appropriate group for the UI tool selector
+- **Tool count assertions** in tests (e.g., `test_mcp_tools.py`) — must be updated to reflect the new count
+
+### 11. Review against other coding standards
+
+Check the diff for violations of the standards listed above:
+- Python files at project root (not allowed)
+- Missing async patterns (should be async throughout)
+- Pydantic models for request/response
+- Line length > 120 chars
+- New features/code beyond what was asked (over-engineering)
+- Unnecessary error handling for impossible scenarios
+- Premature abstractions or speculative helpers
+- Backwards-compatibility hacks (unused vars, re-exports, "removed" comments)
+
+### 12. Report findings
+
+Present a clear summary organized by severity:
+
+**Must fix** — issues that will break CI or violate hard project rules:
+- Unrelated commits on the branch
+- Lint failures
+- Missing type hints on public functions
+- Raw dict usage for structured data (including internal code)
+- Multi-item tuple returns (including internal code)
+- Missing tests for new endpoints
+- New integration missing tests, CI job, or release-integration.sh entry
+
+**Should fix** — issues that hurt code quality:
+- Dead code / unused imports missed by linter
+- Missing tests for non-trivial utility functions
+- Over-engineering beyond the task scope
+
+**Note** — observations that may or may not need action:
+- API changes that might need client regeneration
+- Patterns that deviate from nearby code style
+
+For each finding, include the file path, line number, and a brief explanation.
+
+Do NOT auto-fix any issues. Report all findings and let the user decide what to address. If there are no findings, confirm the code looks good.

.env.example

@@ -2,7 +2,7 @@
 # Copy this file to .env and fill in your values
 
 # LLM Configuration (Required)
-# Supported providers: openai, groq, ollama, gemini, anthropic, lmstudio, vertexai, minimax
+# Supported providers: openai, groq, ollama, gemini, anthropic, lmstudio, vertexai, minimax, volcano
 HINDSIGHT_API_LLM_PROVIDER=openai
 HINDSIGHT_API_LLM_API_KEY=your-api-key-here
 HINDSIGHT_API_LLM_MODEL=gpt-4o-mini
@@ -20,10 +20,10 @@ HINDSIGHT_API_LLM_BASE_URL=https://api.openai.com/v1
 # HINDSIGHT_API_LLM_VERTEXAI_REGION=us-central1
 # HINDSIGHT_API_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY=/path/to/service-account-key.json  # Optional, uses ADC if not set
 
-# Example: MiniMax configuration (204K context window)
+# Example: MiniMax configuration (1M context window)
 # HINDSIGHT_API_LLM_PROVIDER=minimax
 # HINDSIGHT_API_LLM_API_KEY=your-minimax-api-key
-# HINDSIGHT_API_LLM_MODEL=MiniMax-M2.5
+# HINDSIGHT_API_LLM_MODEL=MiniMax-M2.7
 
 # Example: LM Studio local configuration (Qwen 2.5 32B recommended)
 # HINDSIGHT_API_LLM_PROVIDER=lmstudio
@@ -44,6 +44,7 @@ HINDSIGHT_API_LOG_LEVEL=info
 
 # Database (Optional - uses embedded pg0 by default)
 # HINDSIGHT_API_DATABASE_URL=postgresql://user:pass@host:5432/db
+# HINDSIGHT_API_MIGRATION_DATABASE_URL=  # Direct PostgreSQL URL for migrations (bypasses PgBouncer). Falls back to DATABASE_URL.
 # HINDSIGHT_API_DATABASE_SCHEMA=public  # PostgreSQL schema name (default: public)
 
 # Vector Extension (Optional - uses pgvector by default)

.github/workflows/deploy-docs.yml

@@ -44,5 +44,5 @@ jobs:
     runs-on: ubuntu-latest
     needs: build
     steps:
-      - uses: actions/deploy-pages@v4
+      - uses: actions/deploy-pages@v5
         id: deployment

.github/workflows/release-integration.yml

@@ -0,0 +1,120 @@
+name: Release Integration
+
+on:
+  push:
+    tags:
+      - 'integrations/**'
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write  # for PyPI trusted publishing
+
+    steps:
+    - uses: actions/checkout@v6
+
+    - name: Extract integration info
+      id: info
+      run: |
+        # refs/tags/integrations/litellm/v0.1.0 → integration=litellm, version=0.1.0
+        TAG="${GITHUB_REF#refs/tags/}"
+        INTEGRATION=$(echo "$TAG" | cut -d'/' -f2)
+        VERSION=$(echo "$TAG" | cut -d'/' -f3 | sed 's/^v//')
+        echo "integration=$INTEGRATION" >> $GITHUB_OUTPUT
+        echo "version=$VERSION" >> $GITHUB_OUTPUT
+        echo "tag=$TAG" >> $GITHUB_OUTPUT
+        echo "Integration: $INTEGRATION, Version: $VERSION"
+
+    - name: Detect integration type
+      id: type
+      run: |
+        if [ -f "hindsight-integrations/${{ steps.info.outputs.integration }}/pyproject.toml" ]; then
+          echo "type=python" >> $GITHUB_OUTPUT
+        elif [ -f "hindsight-integrations/${{ steps.info.outputs.integration }}/package.json" ]; then
+          echo "type=typescript" >> $GITHUB_OUTPUT
+        else
+          echo "type=plugin" >> $GITHUB_OUTPUT
+        fi
+
+    # ── Python integrations (litellm, pydantic-ai, crewai) ──────────────────
+
+    - name: Install uv
+      if: steps.type.outputs.type == 'python'
+      uses: astral-sh/setup-uv@v7
+      with:
+        enable-cache: true
+
+    - name: Set up Python
+      if: steps.type.outputs.type == 'python'
+      uses: actions/setup-python@v6
+      with:
+        python-version-file: ".python-version"
+
+    - name: Build Python package
+      if: steps.type.outputs.type == 'python'
+      working-directory: ./hindsight-integrations/${{ steps.info.outputs.integration }}
+      run: uv build --out-dir dist
+
+    - name: Publish Python package to PyPI
+      if: steps.type.outputs.type == 'python'
+      uses: pypa/gh-action-pypi-publish@release/v1
+      with:
+        packages-dir: ./hindsight-integrations/${{ steps.info.outputs.integration }}/dist
+        skip-existing: true
+
+    # ── TypeScript integrations (ai-sdk, chat, openclaw) ────────────────────
+
+    # ── Plugin integrations (claude-code) — no package to publish ───────────
+
+    - name: Plugin release
+      if: steps.type.outputs.type == 'plugin'
+      run: |
+        echo "Plugin integration ${{ steps.info.outputs.integration }} v${{ steps.info.outputs.version }} — no package to publish."
+        echo "Users install via: claude plugin marketplace add vectorize-io/hindsight --sparse hindsight-integrations"
+
+    # ── TypeScript integrations (ai-sdk, chat, openclaw) ────────────────────
+
+    - name: Set up Node.js
+      if: steps.type.outputs.type == 'typescript'
+      uses: actions/setup-node@v6
+      with:
+        node-version: '22'
+        registry-url: 'https://registry.npmjs.org'
+
+    # Guard: fail fast if the integration's lockfile resolves any dep from a
+    # monorepo workspace (link=true) or a relative file path. The release
+    # runner has no pre-built workspace `dist/` so `npm run build` would
+    # later fail at tsc with "Cannot find module". See:
+    # https://github.com/vectorize-io/hindsight/issues/… (0.6.0 openclaw retry)
+    - name: Check integration lockfile
+      if: steps.type.outputs.type == 'typescript'
+      run: ./scripts/check-integration-lockfiles.sh
+
+    - name: Install dependencies
+      if: steps.type.outputs.type == 'typescript'
+      working-directory: ./hindsight-integrations/${{ steps.info.outputs.integration }}
+      run: npm ci
+
+    - name: Build TypeScript package
+      if: steps.type.outputs.type == 'typescript'
+      working-directory: ./hindsight-integrations/${{ steps.info.outputs.integration }}
+      run: npm run build
+
+    - name: Publish TypeScript package to npm
+      if: steps.type.outputs.type == 'typescript'
+      working-directory: ./hindsight-integrations/${{ steps.info.outputs.integration }}
+      run: |
+        set +e
+        OUTPUT=$(npm publish --access public 2>&1)
+        EXIT_CODE=$?
+        echo "$OUTPUT"
+        if [ $EXIT_CODE -ne 0 ]; then
+          if echo "$OUTPUT" | grep -q "cannot publish over"; then
+            echo "Package version already published, skipping..."
+            exit 0
+          fi
+          exit $EXIT_CODE
+        fi
+      env:
+        NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}

.github/workflows/release.yml

@@ -21,7 +21,7 @@ jobs:
         enable-cache: true
 
     - name: Set up Python
-      uses: actions/setup-python@v5
+      uses: actions/setup-python@v6
       with:
         python-version-file: ".python-version"
 
@@ -46,22 +46,10 @@ jobs:
       working-directory: ./hindsight-all-slim
       run: uv build --out-dir dist
 
-    - name: Build hindsight-litellm
-      working-directory: ./hindsight-integrations/litellm
-      run: uv build --out-dir dist
-
     - name: Build hindsight-embed
       working-directory: ./hindsight-embed
       run: uv build --out-dir dist
 
-    - name: Build hindsight-crewai
-      working-directory: ./hindsight-integrations/crewai
-      run: uv build --out-dir dist
-
-    - name: Build hindsight-pydantic-ai
-      working-directory: ./hindsight-integrations/pydantic-ai
-      run: uv build --out-dir dist
-
     # Publish in order (client and api-slim first, then api/all wrappers which depend on them)
     - name: Publish hindsight-client to PyPI
       uses: pypa/gh-action-pypi-publish@release/v1
@@ -93,30 +81,12 @@ jobs:
         packages-dir: ./hindsight-all-slim/dist
         skip-existing: true
 
-    - name: Publish hindsight-litellm to PyPI
-      uses: pypa/gh-action-pypi-publish@release/v1
-      with:
-        packages-dir: ./hindsight-integrations/litellm/dist
-        skip-existing: true
-
     - name: Publish hindsight-embed to PyPI
       uses: pypa/gh-action-pypi-publish@release/v1
       with:
         packages-dir: ./hindsight-embed/dist
         skip-existing: true
 
-    - name: Publish hindsight-crewai to PyPI
-      uses: pypa/gh-action-pypi-publish@release/v1
-      with:
-        packages-dir: ./hindsight-integrations/crewai/dist
-        skip-existing: true
-
-    - name: Publish hindsight-pydantic-ai to PyPI
-      uses: pypa/gh-action-pypi-publish@release/v1
-      with:
-        packages-dir: ./hindsight-integrations/pydantic-ai/dist
-        skip-existing: true
-
     # Upload artifacts for GitHub release
     - name: Upload artifacts
       uses: actions/upload-artifact@v7
@@ -128,10 +98,7 @@ jobs:
           hindsight-api/dist/*
           hindsight-all/dist/*
           hindsight-all-slim/dist/*
-          hindsight-integrations/litellm/dist/*
           hindsight-embed/dist/*
-          hindsight-integrations/crewai/dist/*
-          hindsight-integrations/pydantic-ai/dist/*
         retention-days: 1
 
   release-typescript-client:
@@ -183,7 +150,7 @@ jobs:
         path: hindsight-clients/typescript/*.tgz
         retention-days: 1
 
-  release-openclaw-integration:
+  release-hindsight-all-npm:
     runs-on: ubuntu-latest
     environment: npm
 
@@ -195,17 +162,17 @@ jobs:
       with:
         node-version: '22'
         registry-url: 'https://registry.npmjs.org'
+        cache: 'npm'
+        cache-dependency-path: package-lock.json
 
     - name: Install dependencies
-      working-directory: ./hindsight-integrations/openclaw
-      run: npm ci
+      run: npm ci --workspace=hindsight-all-npm
 
     - name: Build
-      working-directory: ./hindsight-integrations/openclaw
-      run: npm run build
+      run: npm run build --workspace=hindsight-all-npm
 
     - name: Publish to npm
-      working-directory: ./hindsight-integrations/openclaw
+      working-directory: ./hindsight-all-npm
       run: |
         set +e
         OUTPUT=$(npm publish --access public 2>&1)
@@ -222,112 +189,14 @@ jobs:
         NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
 
     - name: Pack for GitHub release
-      working-directory: ./hindsight-integrations/openclaw
+      working-directory: ./hindsight-all-npm
       run: npm pack
 
     - name: Upload artifacts
       uses: actions/upload-artifact@v7
       with:
-        name: openclaw-integration
-        path: hindsight-integrations/openclaw/*.tgz
-        retention-days: 1
-
-  release-ai-sdk-integration:
-    runs-on: ubuntu-latest
-    environment: npm
-
-    steps:
-    - uses: actions/checkout@v6
-
-    - name: Set up Node.js
-      uses: actions/setup-node@v6
-      with:
-        node-version: '22'
-        registry-url: 'https://registry.npmjs.org'
-
-    - name: Install dependencies
-      working-directory: ./hindsight-integrations/ai-sdk
-      run: npm ci
-
-    - name: Build
-      working-directory: ./hindsight-integrations/ai-sdk
-      run: npm run build
-
-    - name: Publish to npm
-      working-directory: ./hindsight-integrations/ai-sdk
-      run: |
-        set +e
-        OUTPUT=$(npm publish --access public 2>&1)
-        EXIT_CODE=$?
-        echo "$OUTPUT"
-        if [ $EXIT_CODE -ne 0 ]; then
-          if echo "$OUTPUT" | grep -q "cannot publish over"; then
-            echo "Package version already published, skipping..."
-            exit 0
-          fi
-          exit $EXIT_CODE
-        fi
-      env:
-        NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
-
-    - name: Pack for GitHub release
-      working-directory: ./hindsight-integrations/ai-sdk
-      run: npm pack
-
-    - name: Upload artifacts
-      uses: actions/upload-artifact@v7
-      with:
-        name: ai-sdk-integration
-        path: hindsight-integrations/ai-sdk/*.tgz
-        retention-days: 1
-
-  release-chat-integration:
-    runs-on: ubuntu-latest
-    environment: npm
-
-    steps:
-    - uses: actions/checkout@v6
-
-    - name: Set up Node.js
-      uses: actions/setup-node@v6
-      with:
-        node-version: '22'
-        registry-url: 'https://registry.npmjs.org'
-
-    - name: Install dependencies
-      working-directory: ./hindsight-integrations/chat
-      run: npm ci
-
-    - name: Build
-      working-directory: ./hindsight-integrations/chat
-      run: npm run build
-
-    - name: Publish to npm
-      working-directory: ./hindsight-integrations/chat
-      run: |
-        set +e
-        OUTPUT=$(npm publish --access public 2>&1)
-        EXIT_CODE=$?
-        echo "$OUTPUT"
-        if [ $EXIT_CODE -ne 0 ]; then
-          if echo "$OUTPUT" | grep -q "cannot publish over"; then
-            echo "Package version already published, skipping..."
-            exit 0
-          fi
-          exit $EXIT_CODE
-        fi
-      env:
-        NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
-
-    - name: Pack for GitHub release
-      working-directory: ./hindsight-integrations/chat
-      run: npm pack
-
-    - name: Upload artifacts
-      uses: actions/upload-artifact@v7
-      with:
-        name: chat-integration
-        path: hindsight-integrations/chat/*.tgz
+        name: hindsight-all-npm
+        path: hindsight-all-npm/*.tgz
         retention-days: 1
 
   release-control-plane:
@@ -562,7 +431,7 @@ jobs:
     - uses: actions/checkout@v6
 
     - name: Install Helm
-      uses: azure/setup-helm@v4
+      uses: azure/setup-helm@v5
       with:
         version: 'latest'
 
@@ -587,7 +456,7 @@ jobs:
 
   create-github-release:
     runs-on: ubuntu-latest
-    needs: [release-python-packages, release-typescript-client, release-openclaw-integration, release-ai-sdk-integration, release-chat-integration, release-control-plane, release-rust-cli, release-docker-images, release-helm-chart]
+    needs: [release-python-packages, release-typescript-client, release-hindsight-all-npm, release-control-plane, release-rust-cli, release-docker-images, release-helm-chart]
     permissions:
       contents: write
 
@@ -599,61 +468,49 @@ jobs:
       run: echo "VERSION=${GITHUB_REF#refs/tags/v}" >> $GITHUB_OUTPUT
 
     - name: Download Python packages
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: python-packages
         path: ./artifacts/python-packages
 
     - name: Download TypeScript client
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: typescript-client
         path: ./artifacts/typescript-client
 
-    - name: Download OpenClaw Integration
-      uses: actions/download-artifact@v4
-      with:
-        name: openclaw-integration
-        path: ./artifacts/openclaw-integration
-
-    - name: Download AI SDK Integration
-      uses: actions/download-artifact@v4
-      with:
-        name: ai-sdk-integration
-        path: ./artifacts/ai-sdk-integration
-
-    - name: Download Chat Integration
-      uses: actions/download-artifact@v4
-      with:
-        name: chat-integration
-        path: ./artifacts/chat-integration
-
     - name: Download Control Plane
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: control-plane
         path: ./artifacts/control-plane
 
+    - name: Download hindsight-embed npm wrapper
+      uses: actions/download-artifact@v8
+      with:
+        name: hindsight-all-npm
+        path: ./artifacts/hindsight-all-npm
+
     - name: Download Rust CLI (Linux)
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: rust-cli-hindsight-linux-amd64
         path: ./artifacts/rust-cli-linux
 
     - name: Download Rust CLI (macOS Intel)
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: rust-cli-hindsight-darwin-amd64
         path: ./artifacts/rust-cli-darwin-amd64
 
     - name: Download Rust CLI (macOS ARM)
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: rust-cli-hindsight-darwin-arm64
         path: ./artifacts/rust-cli-darwin-arm64
 
     - name: Download Helm chart
-      uses: actions/download-artifact@v4
+      uses: actions/download-artifact@v8
       with:
         name: helm-chart
         path: ./artifacts/helm-chart
@@ -667,17 +524,11 @@ jobs:
         cp artifacts/python-packages/hindsight-api/dist/* release-assets/ || true
         cp artifacts/python-packages/hindsight-all/dist/* release-assets/ || true
         cp artifacts/python-packages/hindsight-all-slim/dist/* release-assets/ || true
-        cp artifacts/python-packages/hindsight-integrations/litellm/dist/* release-assets/ || true
-        cp artifacts/python-packages/hindsight-integrations/pydantic-ai/dist/* release-assets/ || true
         cp artifacts/python-packages/hindsight-embed/dist/* release-assets/ || true
         # TypeScript client
         cp artifacts/typescript-client/*.tgz release-assets/ || true
-        # OpenClaw Integration
-        cp artifacts/openclaw-integration/*.tgz release-assets/ || true
-        # AI SDK Integration
-        cp artifacts/ai-sdk-integration/*.tgz release-assets/ || true
-        # Chat Integration
-        cp artifacts/chat-integration/*.tgz release-assets/ || true
+        # hindsight-embed npm wrapper
+        cp artifacts/hindsight-all-npm/*.tgz release-assets/ || true
         # Control Plane
         cp artifacts/control-plane/*.tgz release-assets/ || true
         # Rust CLI binaries
@@ -689,7 +540,7 @@ jobs:
         ls -la release-assets/
 
     - name: Create GitHub Release
-      uses: softprops/action-gh-release@v2
+      uses: softprops/action-gh-release@v3
       with:
         files: release-assets/*
         generate_release_notes: true

.gitignore

@@ -50,7 +50,8 @@ hindsight-dev/benchmarks/perf/results/
 benchmarks/results/
 hindsight-cli/target
 hindsight-clients/rust/target
-.claude
+.claude/*
+!.claude/skills/
 whats-next.md
 TASK.md
 # Changelog is now tracked in hindsight-docs/src/pages/changelog.md

CLAUDE.md

@@ -11,9 +11,15 @@ Hindsight is an agent memory system that provides long-term memory for AI agents
 
 ## Development Commands
 
+### Local Development (API + UI)
+```bash
+# Start both API server and control plane UI
+./scripts/dev/start.sh
+```
+
 ### API Server (Python/FastAPI)
 ```bash
-# Start API server (loads .env automatically)
+# Start API server only (loads .env automatically)
 ./scripts/dev/start-api.sh
 
 # Run all tests (parallelized with pytest-xdist)
@@ -73,17 +79,16 @@ cd hindsight-control-plane && npm run dev
 
 ### Monorepo Structure
 - **hindsight-api-slim/**: Core FastAPI server with memory engine (Python, uv)
-- **hindsight/**: Embedded Python bundle (hindsight-all package)
 - **hindsight-control-plane/**: Admin UI (Next.js, npm)
 - **hindsight-cli/**: CLI tool (Rust, cargo, uses progenitor for API client)
 - **hindsight-clients/**: Generated SDK clients (Python, TypeScript, Rust)
 - **hindsight-docs/**: Docusaurus documentation site
-- **hindsight-integrations/**: Framework integrations (LiteLLM, OpenAI)
+- **hindsight-integrations/**: Framework integrations (LiteLLM, CrewAI, LangGraph, Pydantic AI, AG2, Claude Code, etc.)
 - **hindsight-dev/**: Development tools and benchmarks
 
 ### Core Engine (hindsight-api-slim/hindsight_api/engine/)
-- `memory_engine.py`: Main orchestrator (~170KB) for retain/recall/reflect operations
-- `llm_wrapper.py`: LLM abstraction supporting OpenAI, Anthropic, Gemini, Groq, MiniMax, Ollama, LM Studio
+- `memory_engine.py`: Main orchestrator for retain/recall/reflect operations
+- `llm_wrapper.py`: LLM abstraction supporting OpenAI, Anthropic, Gemini, VertexAI, Groq, MiniMax, Ollama, LM Studio, LiteLLM, Claude Code
 - `embeddings.py`: Embedding generation (local sentence-transformers or TEI)
 - `cross_encoder.py`: Reranking (local or TEI)
 - `entity_resolver.py`: Entity extraction and normalization
@@ -96,13 +101,13 @@ cd hindsight-control-plane && npm run dev
 
 **search/**: Multi-strategy retrieval
 - `retrieval.py`: Main retrieval orchestrator
-- `graph_retrieval.py`: Entity/relationship graph traversal
-- `mpfp_retrieval.py`: Multi-Path Fact Propagation retrieval
+- `graph_retrieval.py`: Graph retrieval abstract base class
+- `link_expansion_retrieval.py`: Link expansion graph retrieval
 - `fusion.py`: Reciprocal rank fusion for combining results
 - `reranking.py`: Cross-encoder reranking
 
 ### API Layer (hindsight-api-slim/hindsight_api/api/)
-- `http.py`: FastAPI HTTP routers (~80KB) for all REST endpoints
+- `http.py`: FastAPI HTTP routers for all REST endpoints
 - `mcp.py`: Model Context Protocol server implementation
 
 Main operations:
@@ -164,11 +169,17 @@ Key tables: `banks`, `memory_units`, `documents`, `entities`, `entity_links`
 ## Key Conventions
 
 ### Code Quality
+
+**Before writing code, read `.claude/skills/code-review/SKILL.md`** for the full coding standards (Python style, type safety, TypeScript style, general principles).
+
 **Always run the lint script after making Python or TypeScript/Node changes:**
 ```bash
 ./scripts/hooks/lint.sh
 ```
-This runs the same checks as the pre-commit hook (Ruff for Python, ESLint/Prettier for TypeScript).
+
+**After completing any implementation work, run `/code-review`** to verify your changes against project standards (missing tests, dead code, type safety, etc.). Fix any "must fix" issues before considering the task done.
+
+**MANDATORY: Run `/code-review` before pushing code or creating a pull request.** Do not push or create a PR until all "must fix" issues are resolved.
 
 ### Memory Banks
 - Each bank is an isolated memory store (like a "brain" for one user/agent)
@@ -200,48 +211,20 @@ When adding or modifying parameters in the dataplane API (hindsight-api), you mu
    - Update the client type definition in `lib/api.ts`
    - Update any UI components that need to use the new parameter
 
-### Python Style
-- Python 3.11+, type hints required
-- Async throughout (asyncpg, async FastAPI)
-- Pydantic models for request/response
-- Ruff for linting (line-length 120)
-- No Python files at project root - maintain clean directory structure
-- **Never use multi-item tuple return values** - prefer dataclass or Pydantic model for structured returns
+### Adding New Integrations
 
-### Type Safety with Pydantic Models
-**NEVER use raw `dict` types for structured data.** Always use Pydantic models:
-- Use Pydantic `BaseModel` for all data structures passed between functions
-- Add `@field_validator` for type coercion (e.g., ensuring datetimes are timezone-aware)
-- Avoid `dict.get()` patterns - use typed model attributes instead
-- Parse external data (JSON, API responses) into Pydantic models at the boundary
-- This catches type errors at parse time, not deep in business logic
+Every new integration in `hindsight-integrations/` must satisfy all of the following before it can be merged:
 
-```python
-# BAD - error-prone dict access
-def process(data: dict) -> str:
-    return data.get("name", "")  # No validation, silent failures
+1. **Tests are required** — tests must simulate or exercise the external system (mock the framework's interfaces and verify the integration actually calls Hindsight correctly). Pure unit tests of helper functions are not sufficient.
+2. **CI job** — add a test job in `.github/workflows/test.yml` following the existing pattern (e.g., `test-crewai-integration`). The job must build, install deps, and run `uv run pytest tests -v`. Also add the integration to `detect-changes` outputs so it only runs when its files change.
+3. **Release process** — add the integration name to the `VALID_INTEGRATIONS` array in `scripts/release-integration.sh` so it can be released via the standard release workflow.
+4. **Follow project code standards** — Python style, type safety, no raw dicts for structured data, no multi-item tuple returns (see `.claude/skills/code-review/SKILL.md`).
 
-# GOOD - typed and validated
-class UserData(BaseModel):
-    name: str
-    created_at: datetime
+If any of these are missing, the integration is incomplete and must not be pushed or merged.
 
-    @field_validator("created_at", mode="before")
-    @classmethod
-    def ensure_tz_aware(cls, v):
-        if isinstance(v, str):
-            v = datetime.fromisoformat(v.replace("Z", "+00:00"))
-        if v.tzinfo is None:
-            return v.replace(tzinfo=timezone.utc)
-        return v
+### Changelogs
 
-def process(data: UserData) -> str:
-    return data.name  # Type-safe, validated at construction
-```
-
-### TypeScript Style
-- Next.js App Router for control plane
-- Tailwind CSS with shadcn/ui components
+Never add "Unreleased" entries to changelogs (e.g. `hindsight-docs/src/pages/changelog/**`). Changelog entries are written by the release script (`./scripts/release-integration.sh`) when a version is actually cut. If a bug fix or feature needs documenting before release, describe it in the PR/commit — the release tooling will surface it in the published changelog section.
 
 ### Adding New API Configuration Flags
 
@@ -255,17 +238,17 @@ Fields must be categorized as either **hierarchical** (can be overridden per-ten
    - Add `ENV_*` constant for the environment variable name (e.g., `ENV_MY_SETTING = "HINDSIGHT_API_MY_SETTING"`)
    - Add `DEFAULT_*` constant for the default value
    - Add field to `HindsightConfig` dataclass with type annotation
-   - **Mark as hierarchical or static** by adding to `_HIERARCHICAL_FIELDS` set (hierarchical) or leaving it out (static)
+   - **Mark as configurable** by adding to `_CONFIGURABLE_FIELDS` set if the field should be overridable per-tenant/bank via API
    - Add initialization in `from_env()` method
 
    ```python
-   # Hierarchical field (can be overridden per-bank)
-   _HIERARCHICAL_FIELDS = {
+   # Configurable field (can be overridden per-tenant/bank via API)
+   _CONFIGURABLE_FIELDS = {
        ...,
-       "my_setting",  # Add here for hierarchical
+       "my_setting",  # Add here for configurable
    }
 
-   # Static field - just don't add to _HIERARCHICAL_FIELDS
+   # Static field - just don't add to _CONFIGURABLE_FIELDS
    ```
 
 2. **main.py** (`hindsight-api-slim/hindsight_api/main.py`):

README.md

@@ -7,6 +7,7 @@
 [![CI](https://github.com/vectorize-io/hindsight/actions/workflows/release.yml/badge.svg)](https://github.com/vectorize-io/hindsight/actions/workflows/release.yml)
 [![Slack Community](https://img.shields.io/badge/Slack-Join%20Community-4A154B?logo=slack)](https://join.slack.com/t/hindsight-space/shared_invite/zt-3nhbm4w29-LeSJ5Ixi6j8PdiYOCPlOgg)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+[![gitcgr](https://gitcgr.com/badge/vectorize-io/hindsight.svg)](https://gitcgr.com/vectorize-io/hindsight)
 ![PyPI - Downloads](https://img.shields.io/pypi/dm/hindsight-api?label=PyPI)
 ![NPM Downloads](https://img.shields.io/npm/dm/%40vectorize-io%2Fhindsight-client?logoColor=orange&label=NPM&color=blue&link=https%3A%2F%2Fwww.npmjs.com%2Fpackage%2F%40vectorize-io%2Fhindsight-client)
 <br/>

deno.lock

@@ -0,0 +1,139 @@
+{
+  "version": "5",
+  "specifiers": {
+    "jsr:@std/assert@^1.0.17": "1.0.19",
+    "jsr:@std/assert@^1.0.19": "1.0.19",
+    "jsr:@std/expect@*": "1.0.18",
+    "jsr:@std/internal@^1.0.12": "1.0.12",
+    "jsr:@std/path@^1.1.4": "1.1.4",
+    "jsr:@std/testing@*": "1.0.17"
+  },
+  "jsr": {
+    "@std/[email protected]": {
+      "integrity": "eaada96ee120cb980bc47e040f82814d786fe8162ecc53c91d8df60b8755991e",
+      "dependencies": [
+        "jsr:@std/internal"
+      ]
+    },
+    "@std/[email protected]": {
+      "integrity": "8566eab35200466f8609eb7e7aed062ed0db314e9a258d5d201b1b8997ce801a",
+      "dependencies": [
+        "jsr:@std/assert@^1.0.19",
+        "jsr:@std/internal",
+        "jsr:@std/path"
+      ]
+    },
+    "@std/[email protected]": {
+      "integrity": "972a634fd5bc34b242024402972cd5143eac68d8dffaca5eaa4dba30ce17b027"
+    },
+    "@std/[email protected]": {
+      "integrity": "1d2d43f39efb1b42f0b1882a25486647cb851481862dc7313390b2bb044314b5",
+      "dependencies": [
+        "jsr:@std/internal"
+      ]
+    },
+    "@std/[email protected]": {
+      "integrity": "87bdc2700fa98249d48a17cd72413352d3d3680dcfbdb64947fd0982d6bbf681",
+      "dependencies": [
+        "jsr:@std/assert@^1.0.17",
+        "jsr:@std/internal"
+      ]
+    }
+  },
+  "workspace": {
+    "members": {
+      "hindsight-clients/typescript": {
+        "packageJson": {
+          "dependencies": [
+            "npm:@hey-api/[email protected]",
+            "npm:@types/jest@29",
+            "npm:@types/node@20",
+            "npm:jest@29",
+            "npm:ts-jest@29",
+            "npm:tsup@^8.5.1",
+            "npm:typescript@5"
+          ]
+        }
+      },
+      "hindsight-control-plane": {
+        "packageJson": {
+          "dependencies": [
+            "npm:@eslint/eslintrc@^3.3.3",
+            "npm:@eslint/js@^9.39.2",
+            "npm:@radix-ui/react-alert-dialog@^1.1.15",
+            "npm:@radix-ui/react-checkbox@^1.3.3",
+            "npm:@radix-ui/react-dialog@^1.1.15",
+            "npm:@radix-ui/react-dropdown-menu@^2.1.16",
+            "npm:@radix-ui/react-label@^2.1.8",
+            "npm:@radix-ui/react-popover@^1.1.15",
+            "npm:@radix-ui/react-radio-group@^1.3.8",
+            "npm:@radix-ui/react-select@^2.2.6",
+            "npm:@radix-ui/react-slider@^1.3.6",
+            "npm:@radix-ui/react-slot@^1.2.4",
+            "npm:@radix-ui/react-switch@^1.2.6",
+            "npm:@radix-ui/react-tabs@^1.1.13",
+            "npm:@radix-ui/react-tooltip@^1.2.8",
+            "npm:@tailwindcss/postcss@^4.1.17",
+            "npm:@tailwindcss/typography@~0.5.19",
+            "npm:@types/cytoscape@^3.21.9",
+            "npm:@types/node@^24.10.0",
+            "npm:@types/react-dom@^19.2.2",
+            "npm:@types/react@^19.2.2",
+            "npm:autoprefixer@^10.4.21",
+            "npm:class-variance-authority@~0.7.1",
+            "npm:clsx@^2.1.1",
+            "npm:cmdk@^1.1.1",
+            "npm:cytoscape-fcose@^2.2.0",
+            "npm:cytoscape@^3.33.1",
+            "npm:eslint-config-next@^16.0.1",
+            "npm:eslint-plugin-react-hooks@^7.0.1",
+            "npm:eslint-plugin-react@^7.37.5",
+            "npm:eslint@^9.39.1",
+            "npm:[email protected]",
+            "npm:next-themes@~0.4.6",
+            "npm:next@^16.1.6",
+            "npm:postcss@^8.5.6",
+            "npm:prettier@^3.7.4",
+            "npm:react-chrono@^2.9.1",
+            "npm:react-dom@^19.2.0",
+            "npm:react-markdown@^10.1.0",
+            "npm:react18-json-view@~0.2.9",
+            "npm:react@^19.2.0",
+            "npm:recharts@^3.5.1",
+            "npm:remark-gfm@^4.0.1",
+            "npm:sonner@^2.0.7",
+            "npm:tailwind-merge@^3.4.0",
+            "npm:tailwindcss-animate@^1.0.7",
+            "npm:tailwindcss@^4.1.17",
+            "npm:[email protected]",
+            "npm:typescript-eslint@^8.50.0",
+            "npm:typescript@^5.9.3"
+          ]
+        }
+      },
+      "hindsight-docs": {
+        "packageJson": {
+          "dependencies": [
+            "npm:@docusaurus/[email protected]",
+            "npm:@docusaurus/[email protected]",
+            "npm:@docusaurus/[email protected]",
+            "npm:@docusaurus/theme-common@^3.9.2",
+            "npm:@docusaurus/theme-mermaid@^3.9.2",
+            "npm:@docusaurus/[email protected]",
+            "npm:@docusaurus/[email protected]",
+            "npm:@easyops-cn/docusaurus-search-local@~0.52.2",
+            "npm:@mdx-js/react@3",
+            "npm:clsx@2",
+            "npm:prism-react-renderer@^2.3.0",
+            "npm:raw-loader@^4.0.2",
+            "npm:react-dom@19",
+            "npm:react-icons@^5.6.0",
+            "npm:react@19",
+            "npm:redocusaurus@^2.5.0",
+            "npm:typescript@~5.6.2"
+          ]
+        }
+      }
+    }
+  }
+}

docker/standalone/start-all.sh

@@ -1,6 +1,28 @@
 #!/bin/bash
 set -e
 
+# =============================================================================
+# Embedded pg0 data integrity check (#675)
+#
+# When using embedded pg0, check if the data directory has existing PostgreSQL
+# data before starting. If the directory exists but appears empty/corrupt
+# (e.g., missing PG_VERSION file), log a warning. This helps diagnose data
+# loss scenarios where a container restart caused the data directory to be
+# wiped despite a volume mount being present.
+# =============================================================================
+PG0_DATA_DIR="${HOME}/.pg0"
+if [ -d "$PG0_DATA_DIR" ]; then
+    # Look for actual PostgreSQL data directories (pg0 creates subdirs per instance)
+    if compgen -G "$PG0_DATA_DIR"/*/PG_VERSION > /dev/null 2>&1; then
+        echo "✅ Existing pg0 data directory detected at $PG0_DATA_DIR"
+    elif [ "$(ls -A "$PG0_DATA_DIR" 2>/dev/null)" ]; then
+        echo "⚠️  WARNING: pg0 data directory exists at $PG0_DATA_DIR but no PG_VERSION found."
+        echo "   This may indicate data corruption or an incomplete previous shutdown."
+        echo "   If you see all migrations running from scratch after this, your data may have been lost."
+        echo "   See: https://github.com/vectorize-io/hindsight/issues/675"
+    fi
+fi
+
 # Service flags (default to true if not set)
 ENABLE_API="${HINDSIGHT_ENABLE_API:-true}"
 ENABLE_CP="${HINDSIGHT_ENABLE_CP:-true}"
@@ -71,6 +93,63 @@ if [ "${HINDSIGHT_WAIT_FOR_DEPS:-false}" = "true" ]; then
     done
 fi
 
+# =============================================================================
+# Graceful shutdown handler (#675)
+#
+# Docker sends SIGTERM on `docker stop`/`docker restart`. Without a trap, child
+# processes (hindsight-api + pg0, control-plane) are killed abruptly. For the
+# embedded pg0 database this can cause data loss when the data directory is on
+# a Docker volume that gets remounted after restart.
+#
+# The trap forwards SIGTERM to all tracked child PIDs so that:
+#   - hindsight-api receives the signal and can run its shutdown hooks
+#   - pg0 gets a clean PostgreSQL shutdown (checkpoint + WAL flush)
+#   - The control-plane Node.js process exits cleanly
+# =============================================================================
+# Guard against concurrent cleanup (e.g., child crash + SIGTERM arriving together)
+SHUTTING_DOWN=false
+
+cleanup() {
+    if $SHUTTING_DOWN; then return; fi
+    SHUTTING_DOWN=true
+
+    echo ""
+    echo "🛑 Received shutdown signal, stopping services gracefully..."
+    for pid in "${PIDS[@]}"; do
+        if kill -0 "$pid" 2>/dev/null; then
+            kill -TERM "$pid" 2>/dev/null
+        fi
+    done
+    # Give processes time to shut down cleanly (pg0 needs to flush WAL).
+    # NOTE: Docker's default stop_grace_period is 10s. If you use the default,
+    # either set stop_grace_period: 30s in your compose file / docker stop -t 30,
+    # or Docker will SIGKILL the container before this timeout expires.
+    local timeout=30
+    for ((i=1; i<=timeout; i++)); do
+        local all_stopped=true
+        for pid in "${PIDS[@]}"; do
+            if kill -0 "$pid" 2>/dev/null; then
+                all_stopped=false
+                break
+            fi
+        done
+        if $all_stopped; then
+            echo "✅ All services stopped cleanly"
+            exit 0
+        fi
+        sleep 1
+    done
+    # Force kill if still running after timeout
+    echo "⚠️  Timeout reached, forcing shutdown..."
+    for pid in "${PIDS[@]}"; do
+        if kill -0 "$pid" 2>/dev/null; then
+            kill -9 "$pid" 2>/dev/null
+        fi
+    done
+    exit 1
+}
+trap cleanup SIGTERM SIGINT
+
 # Track PIDs for wait
 PIDS=()
 
@@ -111,6 +190,7 @@ fi
 if [ "$ENABLE_CP" = "true" ]; then
     echo "🎛️  Starting Control Plane..."
     cd /app/control-plane
+    export HOSTNAME="${HINDSIGHT_CP_HOSTNAME:-0.0.0.0}"
     PORT="${HINDSIGHT_CP_PORT:-9999}" node server.js &
     CP_PID=$!
     PIDS+=($CP_PID)
@@ -137,8 +217,21 @@ if [ ${#PIDS[@]} -eq 0 ]; then
     exit 1
 fi
 
-# Wait for any process to exit
-wait -n
-
-# Exit with status of first exited process
-exit $?
+# Wait for any process to exit (use wait -n with trap-safe loop)
+while true; do
+    # wait -n returns when any child exits; it also returns on signal delivery
+    # (the trap handler will run and exit, so this loop is just for robustness).
+    # `&& true` prevents `set -e` from killing the script when wait -n returns
+    # non-zero (child exited with error or no backgrounded children remain).
+    wait -n && true
+    # Check if any tracked PID has exited
+    for pid in "${PIDS[@]}"; do
+        if ! kill -0 "$pid" 2>/dev/null; then
+            wait "$pid" 2>/dev/null
+            exit_code=$?
+            echo "⚠️  Service (PID $pid) exited with code $exit_code"
+            # Trigger cleanup for remaining services
+            cleanup
+        fi
+    done
+done

helm/hindsight/Chart.yaml

@@ -2,8 +2,8 @@ apiVersion: v2
 name: hindsight
 description: Hindsight helm chart
 type: application
-version: 0.4.17
-appVersion: "0.4.17"
+version: 0.5.1
+appVersion: "0.5.1"
 keywords:
   - ai
   - memory

helm/hindsight/templates/api-deployment.yaml

@@ -95,6 +95,27 @@ spec:
           {{- toYaml .Values.api.readinessProbe | nindent 10 }}
         resources:
           {{- toYaml .Values.api.resources | nindent 10 }}
+        {{- if or .Values.api.persistence.modelCache.enabled .Values.api.extraVolumeMounts }}
+        volumeMounts:
+        {{- if .Values.api.persistence.modelCache.enabled }}
+        - name: model-cache
+          mountPath: /home/hindsight/.cache
+        {{- end }}
+        {{- with .Values.api.extraVolumeMounts }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+        {{- end }}
+      {{- if or .Values.api.persistence.modelCache.enabled .Values.api.extraVolumes }}
+      volumes:
+      {{- if .Values.api.persistence.modelCache.enabled }}
+      - name: model-cache
+        persistentVolumeClaim:
+          claimName: {{ include "hindsight.fullname" . }}-api-model-cache
+      {{- end }}
+      {{- with .Values.api.extraVolumes }}
+      {{- toYaml . | nindent 6 }}
+      {{- end }}
+      {{- end }}
       {{- with .Values.nodeSelector }}
       nodeSelector:
         {{- toYaml . | nindent 8 }}

helm/hindsight/templates/api-model-cache-pvc.yaml

@@ -0,0 +1,21 @@
+{{- if and .Values.api.enabled .Values.api.persistence.modelCache.enabled }}
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: {{ include "hindsight.fullname" . }}-api-model-cache
+  labels:
+    {{- include "hindsight.api.labels" . | nindent 4 }}
+  {{- with .Values.api.persistence.modelCache.annotations }}
+  annotations:
+    {{- toYaml . | nindent 4 }}
+  {{- end }}
+spec:
+  accessModes:
+    {{- toYaml .Values.api.persistence.modelCache.accessModes | nindent 4 }}
+  {{- if .Values.api.persistence.modelCache.storageClass }}
+  storageClassName: {{ .Values.api.persistence.modelCache.storageClass }}
+  {{- end }}
+  resources:
+    requests:
+      storage: {{ .Values.api.persistence.modelCache.size }}
+{{- end }}

helm/hindsight/templates/worker-statefulset.yaml

@@ -95,6 +95,16 @@ spec:
           {{- toYaml .Values.worker.readinessProbe | nindent 10 }}
         resources:
           {{- toYaml .Values.worker.resources | nindent 10 }}
+        {{- if or .Values.worker.persistence.modelCache.enabled .Values.worker.extraVolumeMounts }}
+        volumeMounts:
+        {{- if .Values.worker.persistence.modelCache.enabled }}
+        - name: model-cache
+          mountPath: /home/hindsight/.cache
+        {{- end }}
+        {{- with .Values.worker.extraVolumeMounts }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+        {{- end }}
       {{- with .Values.nodeSelector }}
       nodeSelector:
         {{- toYaml . | nindent 8 }}
@@ -107,4 +117,26 @@ spec:
       tolerations:
         {{- toYaml . | nindent 8 }}
       {{- end }}
+      {{- with .Values.worker.extraVolumes }}
+      volumes:
+      {{- toYaml . | nindent 6 }}
+      {{- end }}
+  {{- if .Values.worker.persistence.modelCache.enabled }}
+  volumeClaimTemplates:
+  - metadata:
+      name: model-cache
+      {{- with .Values.worker.persistence.modelCache.annotations }}
+      annotations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+    spec:
+      accessModes:
+        {{- toYaml .Values.worker.persistence.modelCache.accessModes | nindent 8 }}
+      {{- if .Values.worker.persistence.modelCache.storageClass }}
+      storageClassName: {{ .Values.worker.persistence.modelCache.storageClass }}
+      {{- end }}
+      resources:
+        requests:
+          storage: {{ .Values.worker.persistence.modelCache.size }}
+  {{- end }}
 {{- end }}

helm/hindsight/values.yaml

@@ -67,6 +67,33 @@ api:
   # Pod affinity/anti-affinity (overrides global affinity for this component)
   # affinity: {}
 
+  # Persistent volume for local model cache (reranker, embeddings)
+  # Models are downloaded to /home/hindsight/.cache on first use.
+  # Without persistence, models are re-downloaded on every pod restart.
+  persistence:
+    modelCache:
+      enabled: false
+      size: 5Gi
+      storageClass: ""
+      accessModes:
+        - ReadWriteOnce
+      annotations: {}
+
+  # Extra volume mounts for the api container
+  # e.g.
+  # extraVolumeMounts:
+  #   - name: my-volume
+  #     mountPath: /mnt/my-volume
+  extraVolumeMounts: []
+
+  # Extra volumes for the api pod
+  # e.g.
+  # extraVolumes:
+  #   - name: my-volume
+  #     configMap:
+  #       name: my-configmap
+  extraVolumes: []
+
   # Environment variables
   env:
     #HINDSIGHT_API_LLM_PROVIDER: "groq"
@@ -140,6 +167,32 @@ worker:
   # Pod affinity/anti-affinity (overrides global affinity for this component)
   # affinity: {}
 
+  # Persistent volume for local model cache (reranker, embeddings)
+  # Uses volumeClaimTemplates since worker is a StatefulSet.
+  persistence:
+    modelCache:
+      enabled: false
+      size: 5Gi
+      storageClass: ""
+      accessModes:
+        - ReadWriteOnce
+      annotations: {}
+
+  # Extra volume mounts for the worker container
+  # e.g.
+  # extraVolumeMounts:
+  #   - name: my-volume
+  #     mountPath: /mnt/my-volume
+  extraVolumeMounts: []
+
+  # Extra volumes for the worker pod
+  # e.g.
+  # extraVolumes:
+  #   - name: my-volume
+  #     configMap:
+  #       name: my-configmap
+  extraVolumes: []
+
   # Secret environment variables (inherited from api.secrets if not specified)
   secrets: {}
 

hindsight-all-npm/.gitignore

@@ -0,0 +1,4 @@
+node_modules
+dist
+*.tgz
+.DS_Store

hindsight-all-npm/README.md

@@ -0,0 +1,80 @@
+# @vectorize-io/hindsight-all
+
+Node.js equivalent of the Python [`hindsight-all`](https://pypi.org/project/hindsight-all/) package — programmatic lifecycle manager for a local Hindsight daemon. Use this when you want to embed Hindsight in a Node application without hand-rolling subprocess management.
+
+This package deliberately does **not** ship an HTTP client. Once the daemon is running, talk to it with [`@vectorize-io/hindsight-client`](https://www.npmjs.com/package/@vectorize-io/hindsight-client) against `server.getBaseUrl()`. The two packages compose — one owns the daemon process, the other owns the HTTP API surface.
+
+## Requirements
+
+- **Node.js >= 22** — uses global `fetch` and `AbortSignal.timeout`.
+- **`uv` / `uvx`** on `PATH` — used to download and run the underlying `hindsight-embed` daemon on first use. Install via <https://docs.astral.sh/uv/>.
+
+## Install
+
+```bash
+npm install @vectorize-io/hindsight-all @vectorize-io/hindsight-client
+```
+
+## Example
+
+```ts
+import { HindsightServer, consoleLogger } from '@vectorize-io/hindsight-all';
+import { HindsightClient } from '@vectorize-io/hindsight-client';
+
+const server = new HindsightServer({
+  profile: 'my-app',
+  port: 9077,
+  env: {
+    HINDSIGHT_API_LLM_PROVIDER: 'anthropic',
+    HINDSIGHT_API_LLM_API_KEY: process.env.ANTHROPIC_API_KEY,
+    HINDSIGHT_API_LLM_MODEL: 'claude-sonnet-4-20250514',
+    HINDSIGHT_EMBED_DAEMON_IDLE_TIMEOUT: '0',
+  },
+  logger: consoleLogger,
+});
+
+await server.start();
+
+const client = new HindsightClient({ baseUrl: server.getBaseUrl() });
+
+await client.retain('user-123', 'User prefers dark mode and concise answers.', {
+  documentId: 'pref-2026-04-01',
+});
+
+const recall = await client.recall('user-123', 'what are the user preferences?');
+console.log(recall.results);
+
+await server.stop();
+```
+
+For a remote Hindsight API, skip `HindsightServer` entirely and just point `HindsightClient` at the remote URL.
+
+## Open config — forward-compatible with new daemon flags
+
+`HindsightServerOptions` is designed so every new environment variable or CLI flag in the underlying Hindsight daemon can be used without waiting for a wrapper release:
+
+- **`env`** accepts an arbitrary `Record<string, string>`. Every entry is exported into the daemon process and written into the profile config via `--env KEY=VALUE`.
+- **`extraProfileCreateArgs`** / **`extraDaemonStartArgs`** append raw args to the respective commands.
+
+## Development against a local checkout
+
+If you're hacking on the Python `hindsight-embed` package in the same monorepo, point the server at the local path — it'll use `uv run --directory <path>` instead of `uvx`:
+
+```ts
+new HindsightServer({
+  embedPackagePath: '/path/to/hindsight-embed',
+  // ...
+});
+```
+
+## API surface
+
+- `HindsightServer` — daemon lifecycle (`start`, `stop`, `checkHealth`, `getBaseUrl`, `getProfile`).
+- `Logger` interface plus `silentLogger` (default) and `consoleLogger` helpers.
+- `getEmbedCommand(opts)` — low-level helper that returns the `[cmd, ...args]` tuple used to invoke the underlying Python CLI.
+
+For memory operations (retain, recall, reflect, bank management, stats) use [`@vectorize-io/hindsight-client`](https://www.npmjs.com/package/@vectorize-io/hindsight-client).
+
+## License
+
+MIT

hindsight-all-npm/package.json

@@ -0,0 +1,57 @@
+{
+  "name": "@vectorize-io/hindsight-all",
+  "version": "0.5.1",
+  "description": "Node.js programmatic lifecycle manager for Hindsight — embeds a local hindsight daemon in a Node application. Pair with @vectorize-io/hindsight-client for memory operations.",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "type": "module",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    }
+  },
+  "keywords": [
+    "hindsight",
+    "hindsight-all",
+    "memory",
+    "ai",
+    "agent",
+    "long-term-memory",
+    "llm",
+    "embedded-server"
+  ],
+  "author": "Vectorize <[email protected]>",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/vectorize-io/hindsight.git",
+    "directory": "hindsight-all-npm"
+  },
+  "files": [
+    "dist",
+    "README.md"
+  ],
+  "scripts": {
+    "build": "tsup",
+    "dev": "tsup --watch",
+    "clean": "rm -rf dist",
+    "test": "vitest run src",
+    "test:watch": "vitest src",
+    "prepublishOnly": "npm run clean && npm run build"
+  },
+  "devDependencies": {
+    "@types/node": "^22.0.0",
+    "tsup": "^8.5.1",
+    "typescript": "^5.7.0",
+    "vitest": "^4.1.2"
+  },
+  "engines": {
+    "node": ">=22"
+  },
+  "overrides": {
+    "rollup": "^4.59.0",
+    "picomatch": ">=2.3.2 <3.0.0 || >=4.0.4",
+    "vite": ">=8.0.5"
+  }
+}

hindsight-all-npm/src/command.test.ts

@@ -0,0 +1,32 @@
+import { describe, it, expect } from 'vitest';
+import { getEmbedCommand } from './command.js';
+
+describe('getEmbedCommand', () => {
+  it('defaults to uvx hindsight-embed@latest', () => {
+    expect(getEmbedCommand()).toEqual(['uvx', 'hindsight-embed@latest']);
+  });
+
+  it('honours an explicit version', () => {
+    expect(getEmbedCommand({ embedVersion: '0.5.0' })).toEqual(['uvx', '[email protected]']);
+  });
+
+  it('treats an empty version as latest', () => {
+    expect(getEmbedCommand({ embedVersion: '' })).toEqual(['uvx', 'hindsight-embed@latest']);
+  });
+
+  it('uses uv run --directory when a local path is given', () => {
+    expect(getEmbedCommand({ embedPackagePath: '/abs/path' })).toEqual([
+      'uv',
+      'run',
+      '--directory',
+      '/abs/path',
+      'hindsight-embed',
+    ]);
+  });
+
+  it('local path takes precedence over version', () => {
+    expect(
+      getEmbedCommand({ embedPackagePath: '/abs/path', embedVersion: '0.5.0' }),
+    ).toEqual(['uv', 'run', '--directory', '/abs/path', 'hindsight-embed']);
+  });
+});

hindsight-all-npm/src/command.ts

@@ -0,0 +1,25 @@
+/**
+ * Resolve the command that invokes the `hindsight-embed` Python CLI.
+ *
+ * - If `embedPackagePath` is set, runs the package from a local checkout via
+ *   `uv run --directory <path> hindsight-embed`. Used for in-repo development.
+ * - Otherwise runs it via `uvx hindsight-embed@<version>` so no global install
+ *   is required.
+ *
+ * Returns the argv as `[command, ...baseArgs]` suitable for `spawn()` /
+ * `execFile()` (never shell-interpolated).
+ */
+export interface EmbedCommandOptions {
+  /** Version spec passed to uvx (e.g. "latest", "0.5.0"). Default: "latest". */
+  embedVersion?: string;
+  /** Local checkout path. When set, overrides `embedVersion` and uses `uv run`. */
+  embedPackagePath?: string;
+}
+
+export function getEmbedCommand(opts: EmbedCommandOptions = {}): string[] {
+  if (opts.embedPackagePath) {
+    return ['uv', 'run', '--directory', opts.embedPackagePath, 'hindsight-embed'];
+  }
+  const version = opts.embedVersion && opts.embedVersion.length > 0 ? opts.embedVersion : 'latest';
+  return ['uvx', `hindsight-embed@${version}`];
+}

hindsight-all-npm/src/index.ts

@@ -0,0 +1,7 @@
+export { HindsightServer } from './server.js';
+export { getEmbedCommand } from './command.js';
+export { silentLogger, consoleLogger } from './logger.js';
+
+export type { Logger } from './logger.js';
+export type { EmbedCommandOptions } from './command.js';
+export type { HindsightServerOptions } from './types.js';

hindsight-all-npm/src/logger.ts

@@ -0,0 +1,29 @@
+/**
+ * Pluggable logger interface.
+ *
+ * This package does not own any logging infrastructure — consumers inject
+ * whatever they want (console, pino, openclaw's logger, a no-op). The default
+ * is silent so embedding this package never adds noise to an unrelated app.
+ */
+export interface Logger {
+  debug(msg: string): void;
+  info(msg: string): void;
+  warn(msg: string): void;
+  error(msg: string): void;
+}
+
+/** Logger that drops every call. Used when no logger is passed. */
+export const silentLogger: Logger = {
+  debug: () => {},
+  info: () => {},
+  warn: () => {},
+  error: () => {},
+};
+
+/** Logger that writes to the standard console. Handy for CLIs and tests. */
+export const consoleLogger: Logger = {
+  debug: (msg) => console.debug(msg),
+  info: (msg) => console.log(msg),
+  warn: (msg) => console.warn(msg),
+  error: (msg) => console.error(msg),
+};

hindsight-all-npm/src/server.test.ts

@@ -0,0 +1,35 @@
+import { describe, it, expect } from 'vitest';
+import { HindsightServer } from './server.js';
+
+describe('HindsightServer construction', () => {
+  it('defaults base URL to http://127.0.0.1:8888', () => {
+    const server = new HindsightServer();
+    expect(server.getBaseUrl()).toBe('http://127.0.0.1:8888');
+    expect(server.getProfile()).toBe('default');
+  });
+
+  it('honours custom profile, port, and host', () => {
+    const server = new HindsightServer({ profile: 'app', port: 9077, host: '0.0.0.0' });
+    expect(server.getProfile()).toBe('app');
+    expect(server.getBaseUrl()).toBe('http://0.0.0.0:9077');
+  });
+
+  it('accepts open env pass-through without complaining about unknown keys', () => {
+    const server = new HindsightServer({
+      env: {
+        HINDSIGHT_API_LLM_PROVIDER: 'openai',
+        HINDSIGHT_API_LLM_MODEL: 'gpt-4o-mini',
+        // A field that does not exist today — should still be accepted
+        HINDSIGHT_FUTURE_FLAG: 'enabled',
+      },
+    });
+    expect(server).toBeInstanceOf(HindsightServer);
+  });
+
+  it('exposes checkHealth that returns false when no daemon is running', async () => {
+    // Random high port that nothing is listening on.
+    const server = new HindsightServer({ port: 1, readyTimeoutMs: 100 });
+    const healthy = await server.checkHealth();
+    expect(healthy).toBe(false);
+  });
+});

hindsight-all-npm/src/server.ts

@@ -0,0 +1,322 @@
+import { spawn } from 'child_process';
+import { getEmbedCommand } from './command.js';
+import { silentLogger } from './logger.js';
+import type { Logger } from './logger.js';
+import type { HindsightServerOptions } from './types.js';
+
+const DEFAULT_PORT = 8888;
+const DEFAULT_HOST = '127.0.0.1';
+const DEFAULT_PROFILE = 'default';
+const DEFAULT_READY_TIMEOUT_MS = 30_000;
+const DEFAULT_READY_POLL_INTERVAL_MS = 1_000;
+
+/**
+ * Manages the lifecycle of a local Hindsight daemon from a Node.js process.
+ *
+ * On {@link start}, this class:
+ *   1. Resolves the `hindsight-embed` command (via `uvx` or a local `uv run`).
+ *   2. Runs `profile create <name> --merge --port <port> [--env K=V ...]`
+ *      with every entry in {@link HindsightServerOptions.env} forwarded as
+ *      an `--env` flag.
+ *   3. Runs `daemon --profile <name> start` and waits for the start command
+ *      to exit.
+ *   4. Polls `http://host:port/health` until it returns `200` or the
+ *      `readyTimeoutMs` budget is exhausted.
+ *
+ * On {@link stop}, it runs `daemon --profile <name> stop` and returns once
+ * the command exits (or after a short grace period).
+ *
+ * This is the Node.js equivalent of the Python `hindsight-all` package's
+ * `HindsightServer`: a thin programmatic lifecycle wrapper around the
+ * Hindsight daemon. It does NOT ship an HTTP client — once `start()`
+ * resolves, use `@vectorize-io/hindsight-client` against `getBaseUrl()` for
+ * retain / recall / reflect.
+ *
+ * The class is deliberately transparent about the daemon: new CLI flags or
+ * environment variables never require a code change here — callers can pass
+ * them via `env`, `extraProfileCreateArgs`, or `extraDaemonStartArgs`.
+ */
+export class HindsightServer {
+  private readonly profile: string;
+  private readonly port: number;
+  private readonly host: string;
+  private readonly baseUrl: string;
+  private readonly embedVersion: string | undefined;
+  private readonly embedPackagePath: string | undefined;
+  private readonly userEnv: Record<string, string | undefined>;
+  private readonly extraProfileCreateArgs: string[];
+  private readonly extraDaemonStartArgs: string[];
+  private readonly platformCpuWorkaround: boolean;
+  private readonly readyTimeoutMs: number;
+  private readonly readyPollIntervalMs: number;
+  private readonly logger: Logger;
+
+  constructor(opts: HindsightServerOptions = {}) {
+    this.profile = opts.profile ?? DEFAULT_PROFILE;
+    this.port = opts.port ?? DEFAULT_PORT;
+    this.host = opts.host ?? DEFAULT_HOST;
+    this.baseUrl = `http://${this.host}:${this.port}`;
+    this.embedVersion = opts.embedVersion;
+    this.embedPackagePath = opts.embedPackagePath;
+    this.userEnv = opts.env ?? {};
+    this.extraProfileCreateArgs = opts.extraProfileCreateArgs ?? [];
+    this.extraDaemonStartArgs = opts.extraDaemonStartArgs ?? [];
+    this.platformCpuWorkaround = opts.platformCpuWorkaround ?? (process.platform === 'darwin');
+    this.readyTimeoutMs = opts.readyTimeoutMs ?? DEFAULT_READY_TIMEOUT_MS;
+    this.readyPollIntervalMs = opts.readyPollIntervalMs ?? DEFAULT_READY_POLL_INTERVAL_MS;
+    this.logger = opts.logger ?? silentLogger;
+  }
+
+  /** The base URL the daemon listens on (`http://host:port`). */
+  getBaseUrl(): string {
+    return this.baseUrl;
+  }
+
+  /** The profile name this server operates on. */
+  getProfile(): string {
+    return this.profile;
+  }
+
+  /**
+   * Ensure the daemon is configured and running. Idempotent — the underlying
+   * `profile create --merge` and `daemon start` commands tolerate re-runs.
+   */
+  async start(): Promise<void> {
+    this.logger.info(`[hindsight] starting daemon for profile "${this.profile}"`);
+
+    const env = this.buildEnv();
+    await this.configureProfile(env);
+    await this.startDaemon(env);
+    await this.waitForReady();
+
+    this.logger.info(`[hindsight] daemon ready at ${this.baseUrl}`);
+  }
+
+  /** Stop the daemon. Never throws — logs and resolves even on failure. */
+  async stop(): Promise<void> {
+    this.logger.info(`[hindsight] stopping daemon for profile "${this.profile}"`);
+
+    const [cmd, ...baseArgs] = getEmbedCommand({
+      embedVersion: this.embedVersion,
+      embedPackagePath: this.embedPackagePath,
+    });
+    const args = [...baseArgs, 'daemon', '--profile', this.profile, 'stop'];
+
+    const child = spawn(cmd, args, { stdio: 'pipe' });
+    this.pipeOutput(child, 'daemon.stop');
+
+    await new Promise<void>((resolve) => {
+      const timeout = setTimeout(() => {
+        this.logger.warn(`[hindsight] daemon stop timed out after 5s`);
+        resolve();
+      }, 5_000);
+      child.on('exit', () => {
+        clearTimeout(timeout);
+        this.logger.info(`[hindsight] daemon stopped`);
+        resolve();
+      });
+      child.on('error', (err) => {
+        clearTimeout(timeout);
+        this.logger.warn(`[hindsight] error stopping daemon: ${err.message}`);
+        resolve();
+      });
+    });
+  }
+
+  /** Probe `/health` once with a short timeout. */
+  async checkHealth(): Promise<boolean> {
+    try {
+      const res = await fetch(`${this.baseUrl}/health`, {
+        signal: AbortSignal.timeout(2_000),
+      });
+      return res.ok;
+    } catch {
+      return false;
+    }
+  }
+
+  // -------------------------------------------------------------------------
+  // Internal
+  // -------------------------------------------------------------------------
+
+  /**
+   * Merge the process env, the caller-supplied `env`, and (on macOS) the
+   * embeddings CPU workaround. Caller-supplied values always win over the
+   * workaround; undefined values are dropped.
+   */
+  private buildEnv(): NodeJS.ProcessEnv {
+    const merged: NodeJS.ProcessEnv = { ...process.env };
+
+    if (this.platformCpuWorkaround && process.platform === 'darwin') {
+      merged['HINDSIGHT_API_EMBEDDINGS_LOCAL_FORCE_CPU'] = '1';
+      merged['HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU'] = '1';
+    }
+
+    for (const [key, value] of Object.entries(this.userEnv)) {
+      if (value !== undefined) {
+        merged[key] = value;
+      }
+    }
+
+    return merged;
+  }
+
+  /**
+   * Run `profile create <name> --merge --port <port> [--env K=V ...]`.
+   * Every entry in the merged env that was passed via {@link userEnv} (or
+   * auto-applied by the CPU workaround) is forwarded as `--env`.
+   */
+  private async configureProfile(env: NodeJS.ProcessEnv): Promise<void> {
+    this.logger.info(`[hindsight] configuring profile "${this.profile}"`);
+
+    const [cmd, ...baseArgs] = getEmbedCommand({
+      embedVersion: this.embedVersion,
+      embedPackagePath: this.embedPackagePath,
+    });
+    const createArgs = [
+      ...baseArgs,
+      'profile',
+      'create',
+      this.profile,
+      '--merge',
+      '--port',
+      String(this.port),
+    ];
+
+    // Forward every env var that the caller intended for the daemon as --env.
+    // We only forward keys the caller explicitly set (userEnv) plus the CPU
+    // workaround values — not the entire process.env, to avoid leaking random
+    // host state into profile config.
+    const envForProfile = this.collectProfileEnv(env);
+    for (const [key, value] of Object.entries(envForProfile)) {
+      createArgs.push('--env', `${key}=${value}`);
+    }
+
+    createArgs.push(...this.extraProfileCreateArgs);
+
+    await this.runCommand(cmd, createArgs, env, 'profile.create');
+  }
+
+  /** Collect only the env vars that should be written into the profile file. */
+  private collectProfileEnv(env: NodeJS.ProcessEnv): Record<string, string> {
+    const out: Record<string, string> = {};
+
+    // 1. User-supplied env — always forwarded.
+    for (const [key, value] of Object.entries(this.userEnv)) {
+      if (value !== undefined) {
+        out[key] = value;
+      }
+    }
+
+    // 2. CPU workaround — only if auto-applied and not already overridden.
+    if (this.platformCpuWorkaround && process.platform === 'darwin') {
+      const cpuKeys = [
+        'HINDSIGHT_API_EMBEDDINGS_LOCAL_FORCE_CPU',
+        'HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU',
+      ];
+      for (const key of cpuKeys) {
+        if (!(key in out) && env[key] !== undefined) {
+          out[key] = env[key] as string;
+        }
+      }
+    }
+
+    return out;
+  }
+
+  private async startDaemon(env: NodeJS.ProcessEnv): Promise<void> {
+    const [cmd, ...baseArgs] = getEmbedCommand({
+      embedVersion: this.embedVersion,
+      embedPackagePath: this.embedPackagePath,
+    });
+    const args = [
+      ...baseArgs,
+      'daemon',
+      '--profile',
+      this.profile,
+      'start',
+      ...this.extraDaemonStartArgs,
+    ];
+
+    await this.runCommand(cmd, args, env, 'daemon.start');
+  }
+
+  /**
+   * Spawn `cmd` with `args`, pipe its output through the logger, and resolve
+   * once it exits with code 0. Rejects on non-zero exit or spawn error.
+   */
+  private async runCommand(
+    cmd: string,
+    args: string[],
+    env: NodeJS.ProcessEnv,
+    label: string,
+  ): Promise<void> {
+    const child = spawn(cmd, args, { stdio: 'pipe', env });
+    let output = '';
+    child.stdout?.on('data', (data: Buffer) => {
+      const text = data.toString();
+      output += text;
+      for (const line of text.trimEnd().split('\n')) {
+        if (line) this.logger.info(`[hindsight:${label}] ${line}`);
+      }
+    });
+    child.stderr?.on('data', (data: Buffer) => {
+      const text = data.toString();
+      output += text;
+      for (const line of text.trimEnd().split('\n')) {
+        if (line) this.logger.warn(`[hindsight:${label}] ${line}`);
+      }
+    });
+
+    await new Promise<void>((resolve, reject) => {
+      child.on('exit', (code) => {
+        if (code === 0) {
+          resolve();
+        } else {
+          reject(new Error(`${label} failed with code ${code}: ${output.trim()}`));
+        }
+      });
+      child.on('error', (err) => {
+        reject(new Error(`${label} failed to spawn: ${err.message}`, { cause: err }));
+      });
+    });
+  }
+
+  /** Stream a spawned child's stdout/stderr through the logger without blocking. */
+  private pipeOutput(child: ReturnType<typeof spawn>, label: string): void {
+    child.stdout?.on('data', (data: Buffer) => {
+      for (const line of data.toString().trimEnd().split('\n')) {
+        if (line) this.logger.info(`[hindsight:${label}] ${line}`);
+      }
+    });
+    child.stderr?.on('data', (data: Buffer) => {
+      for (const line of data.toString().trimEnd().split('\n')) {
+        if (line) this.logger.warn(`[hindsight:${label}] ${line}`);
+      }
+    });
+  }
+
+  /** Poll `/health` until it succeeds or `readyTimeoutMs` elapses. */
+  private async waitForReady(): Promise<void> {
+    const deadline = Date.now() + this.readyTimeoutMs;
+    let attempt = 0;
+    while (Date.now() < deadline) {
+      attempt++;
+      try {
+        const res = await fetch(`${this.baseUrl}/health`, {
+          signal: AbortSignal.timeout(this.readyPollIntervalMs),
+        });
+        if (res.ok) {
+          this.logger.debug(`[hindsight] health check passed (attempt ${attempt})`);
+          return;
+        }
+      } catch {
+        // expected while the daemon is still booting
+      }
+      await new Promise((resolve) => setTimeout(resolve, this.readyPollIntervalMs));
+    }
+    throw new Error(
+      `Hindsight daemon did not become ready within ${this.readyTimeoutMs}ms at ${this.baseUrl}`,
+    );
+  }
+}

hindsight-all-npm/src/types.ts

@@ -0,0 +1,54 @@
+import type { Logger } from './logger.js';
+
+/**
+ * Options for {@link HindsightServer}.
+ *
+ * The server is intentionally thin and pass-through: anything configurable
+ * on the daemon side (env vars or CLI flags) can be set here without needing
+ * a new dedicated option. Use {@link env} for `HINDSIGHT_*` / `OPENAI_API_KEY` /
+ * custom provider settings, and the two `extra*` arrays to append raw CLI
+ * args to `profile create` or `daemon start`.
+ *
+ * For talking to the daemon after `start()`, use `@vectorize-io/hindsight-client`
+ * against `server.getBaseUrl()`. This package does not ship its own HTTP
+ * client.
+ */
+export interface HindsightServerOptions {
+  /** Profile name used for `--profile <name>` on every sub-command. Default: `"default"`. */
+  profile?: string;
+  /** TCP port the daemon listens on. Default: `8888`. */
+  port?: number;
+  /** Hostname the daemon binds to (for health checks). Default: `127.0.0.1`. */
+  host?: string;
+  /** Version of the underlying `hindsight-embed` PyPI package to run via `uvx`. Default: `"latest"`. */
+  embedVersion?: string;
+  /** Local path to a `hindsight-embed` checkout — takes precedence over `embedVersion`. */
+  embedPackagePath?: string;
+  /**
+   * Environment variables passed to the daemon process AND written into the
+   * profile via repeated `--env KEY=VALUE` flags. This is the preferred way
+   * to surface any `HINDSIGHT_API_*` / `HINDSIGHT_EMBED_*` setting — adding a
+   * new daemon env var never requires a wrapper update.
+   *
+   * Values of `undefined` are dropped (so you can spread conditionally).
+   */
+  env?: Record<string, string | undefined>;
+  /** Extra args appended verbatim to `hindsight-embed profile create <name> --merge ...`. */
+  extraProfileCreateArgs?: string[];
+  /** Extra args appended verbatim to `hindsight-embed daemon --profile <name> start ...`. */
+  extraDaemonStartArgs?: string[];
+  /**
+   * On macOS, automatically set
+   * `HINDSIGHT_API_EMBEDDINGS_LOCAL_FORCE_CPU=1` and
+   * `HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU=1` to avoid Metal/MPS crashes in
+   * daemon mode. Default: `true` on `darwin`, ignored elsewhere. Any value set
+   * explicitly in {@link env} wins over the auto-applied value.
+   */
+  platformCpuWorkaround?: boolean;
+  /** Max time (ms) to wait for `/health` to return 200. Default: `30_000`. */
+  readyTimeoutMs?: number;
+  /** Polling interval (ms) while waiting for `/health`. Default: `1_000`. */
+  readyPollIntervalMs?: number;
+  /** Optional pluggable logger. Default: silent. */
+  logger?: Logger;
+}

hindsight-all-npm/tsconfig.json

@@ -0,0 +1,18 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ES2022",
+    "lib": ["ES2022"],
+    "moduleResolution": "node",
+    "declaration": true,
+    "outDir": "./dist",
+    "rootDir": "./src",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true,
+    "resolveJsonModule": true
+  },
+  "include": ["src/**/*"],
+  "exclude": ["node_modules", "dist", "src/**/*.test.ts"]
+}

hindsight-all-npm/tsup.config.ts

@@ -0,0 +1,11 @@
+import { defineConfig } from 'tsup';
+
+export default defineConfig({
+  entry: ['src/index.ts'],
+  format: ['esm'],
+  dts: true,
+  outDir: 'dist',
+  clean: true,
+  sourcemap: true,
+  bundle: true,
+});

hindsight-all-npm/vitest.config.ts

@@ -0,0 +1,8 @@
+import { defineConfig } from 'vitest/config';
+
+export default defineConfig({
+  test: {
+    include: ['src/**/*.test.ts'],
+    environment: 'node',
+  },
+});

hindsight-all-slim/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "hindsight-all-slim"
-version = "0.4.17"
+version = "0.5.1"
 description = "Hindsight: Agent Memory That Works Like Human Memory - Slim All-in-One Bundle"
 readme = "README.md"
 requires-python = ">=3.11"

hindsight-all/hindsight/client_wrapper.py

@@ -13,7 +13,10 @@ from hindsight_client import Hindsight
 
 
 class BanksAPI:
-    """Namespace for bank-related operations."""
+    """Namespace for bank-related operations.
+    
+    Provides methods to create, delete, and manage memory banks.
+    """
 
     def __init__(self, client: Hindsight):
         self._client = client
@@ -24,8 +27,18 @@ class BanksAPI:
         name: str | None = None,
         mission: str | None = None,
         disposition: dict[str, Any] | None = None,
-    ):
-        """Create a new bank."""
+    ) -> Any:
+        """Create a new bank.
+        
+        Args:
+            bank_id: Unique identifier for the bank.
+            name: Optional display name for the bank.
+            mission: Optional mission statement for the bank.
+            disposition: Optional disposition configuration dict.
+            
+        Returns:
+            Bank creation response from the API.
+        """
         return self._client.create_bank(
             bank_id=bank_id,
             name=name,
@@ -33,27 +46,57 @@ class BanksAPI:
             disposition=disposition,
         )
 
-    def delete(self, bank_id: str):
-        """Delete a bank."""
+    def delete(self, bank_id: str) -> Any:
+        """Delete a bank.
+        
+        Args:
+            bank_id: The ID of the bank to delete.
+            
+        Returns:
+            Deletion response from the API.
+        """
         return self._client.delete_bank(bank_id=bank_id)
 
-    def set_mission(self, bank_id: str, mission: str):
-        """Set or update the mission for a bank."""
+    def set_mission(self, bank_id: str, mission: str) -> Any:
+        """Set or update the mission for a bank.
+        
+        Args:
+            bank_id: The ID of the bank.
+            mission: The mission statement to set.
+            
+        Returns:
+            API response confirming the update.
+        """
         return self._client.set_mission(bank_id=bank_id, mission=mission)
 
-    def set_disposition(self, bank_id: str, disposition: dict[str, Any]):
-        """Set or update the disposition for a bank."""
+    def set_disposition(self, bank_id: str, disposition: dict[str, Any]) -> Any:
+        """Set or update the disposition for a bank.
+        
+        Args:
+            bank_id: The ID of the bank.
+            disposition: The disposition configuration dict.
+            
+        Returns:
+            API response confirming the update.
+        """
         return self._client.set_disposition(bank_id=bank_id, disposition=disposition)
 
-    def list(self):
-        """List all banks."""
+    def list(self) -> Any:
+        """List all banks.
+        
+        Returns:
+            List of banks from the API.
+        """
         from hindsight_client.hindsight_client import _run_async
 
         return _run_async(self._client._banks_api.list_banks())
 
 
 class MentalModelsAPI:
-    """Namespace for mental model operations."""
+    """Namespace for mental model operations.
+    
+    Mental models are reusable knowledge structures that guide agent behavior.
+    """
 
     def __init__(self, client: Hindsight):
         self._client = client
@@ -64,8 +107,18 @@ class MentalModelsAPI:
         name: str,
         content: str,
         tags: list[str] | None = None,
-    ):
-        """Create a new mental model."""
+    ) -> Any:
+        """Create a new mental model.
+        
+        Args:
+            bank_id: The ID of the bank to add the model to.
+            name: Name for the mental model.
+            content: The content/instructions for the mental model.
+            tags: Optional list of tags for categorization.
+            
+        Returns:
+            Creation response from the API.
+        """
         return self._client.create_mental_model(
             bank_id=bank_id,
             name=name,
@@ -73,16 +126,40 @@ class MentalModelsAPI:
             tags=tags,
         )
 
-    def list(self, bank_id: str, tags: list[str] | None = None):
-        """List all mental models for a bank."""
+    def list(self, bank_id: str, tags: list[str] | None = None) -> Any:
+        """List all mental models for a bank.
+        
+        Args:
+            bank_id: The ID of the bank.
+            tags: Optional filter by tags.
+            
+        Returns:
+            List of mental models.
+        """
         return self._client.list_mental_models(bank_id=bank_id, tags=tags)
 
-    def get(self, bank_id: str, mental_model_id: str):
-        """Get a specific mental model."""
+    def get(self, bank_id: str, mental_model_id: str) -> Any:
+        """Get a specific mental model.
+        
+        Args:
+            bank_id: The ID of the bank.
+            mental_model_id: The ID of the mental model.
+            
+        Returns:
+            The mental model details.
+        """
         return self._client.get_mental_model(bank_id=bank_id, mental_model_id=mental_model_id)
 
-    def refresh(self, bank_id: str, mental_model_id: str):
-        """Refresh a mental model."""
+    def refresh(self, bank_id: str, mental_model_id: str) -> Any:
+        """Refresh a mental model.
+        
+        Args:
+            bank_id: The ID of the bank.
+            mental_model_id: The ID of the mental model to refresh.
+            
+        Returns:
+            Refresh response from the API.
+        """
         return self._client.refresh_mental_model(bank_id=bank_id, mental_model_id=mental_model_id)
 
     def update(
@@ -92,8 +169,19 @@ class MentalModelsAPI:
         name: str | None = None,
         content: str | None = None,
         tags: list[str] | None = None,
-    ):
-        """Update a mental model."""
+    ) -> Any:
+        """Update a mental model.
+        
+        Args:
+            bank_id: The ID of the bank.
+            mental_model_id: The ID of the mental model to update.
+            name: Optional new name.
+            content: Optional new content.
+            tags: Optional new tags list.
+            
+        Returns:
+            Update response from the API.
+        """
         return self._client.update_mental_model(
             bank_id=bank_id,
             mental_model_id=mental_model_id,
@@ -102,13 +190,24 @@ class MentalModelsAPI:
             tags=tags,
         )
 
-    def delete(self, bank_id: str, mental_model_id: str):
-        """Delete a mental model."""
+    def delete(self, bank_id: str, mental_model_id: str) -> Any:
+        """Delete a mental model.
+        
+        Args:
+            bank_id: The ID of the bank.
+            mental_model_id: The ID of the mental model to delete.
+            
+        Returns:
+            Deletion response from the API.
+        """
         return self._client.delete_mental_model(bank_id=bank_id, mental_model_id=mental_model_id)
 
 
 class DirectivesAPI:
-    """Namespace for directive operations."""
+    """Namespace for directive operations.
+    
+    Directives are explicit instructions that guide agent behavior.
+    """
 
     def __init__(self, client: Hindsight):
         self._client = client
@@ -119,8 +218,18 @@ class DirectivesAPI:
         name: str,
         content: str,
         tags: list[str] | None = None,
-    ):
-        """Create a new directive."""
+    ) -> Any:
+        """Create a new directive.
+        
+        Args:
+            bank_id: The ID of the bank to add the directive to.
+            name: Name for the directive.
+            content: The directive content/instructions.
+            tags: Optional list of tags for categorization.
+            
+        Returns:
+            Creation response from the API.
+        """
         return self._client.create_directive(
             bank_id=bank_id,
             name=name,
@@ -128,12 +237,28 @@ class DirectivesAPI:
             tags=tags,
         )
 
-    def list(self, bank_id: str, tags: list[str] | None = None):
-        """List all directives for a bank."""
+    def list(self, bank_id: str, tags: list[str] | None = None) -> Any:
+        """List all directives for a bank.
+        
+        Args:
+            bank_id: The ID of the bank.
+            tags: Optional filter by tags.
+            
+        Returns:
+            List of directives.
+        """
         return self._client.list_directives(bank_id=bank_id, tags=tags)
 
-    def get(self, bank_id: str, directive_id: str):
-        """Get a specific directive."""
+    def get(self, bank_id: str, directive_id: str) -> Any:
+        """Get a specific directive.
+        
+        Args:
+            bank_id: The ID of the bank.
+            directive_id: The ID of the directive.
+            
+        Returns:
+            The directive details.
+        """
         return self._client.get_directive(bank_id=bank_id, directive_id=directive_id)
 
     def update(
@@ -143,8 +268,19 @@ class DirectivesAPI:
         name: str | None = None,
         content: str | None = None,
         tags: list[str] | None = None,
-    ):
-        """Update a directive."""
+    ) -> Any:
+        """Update a directive.
+        
+        Args:
+            bank_id: The ID of the bank.
+            directive_id: The ID of the directive to update.
+            name: Optional new name.
+            content: Optional new content.
+            tags: Optional new tags list.
+            
+        Returns:
+            Update response from the API.
+        """
         return self._client.update_directive(
             bank_id=bank_id,
             directive_id=directive_id,
@@ -153,13 +289,24 @@ class DirectivesAPI:
             tags=tags,
         )
 
-    def delete(self, bank_id: str, directive_id: str):
-        """Delete a directive."""
+    def delete(self, bank_id: str, directive_id: str) -> Any:
+        """Delete a directive.
+        
+        Args:
+            bank_id: The ID of the bank.
+            directive_id: The ID of the directive to delete.
+            
+        Returns:
+            Deletion response from the API.
+        """
         return self._client.delete_directive(bank_id=bank_id, directive_id=directive_id)
 
 
 class MemoriesAPI:
-    """Namespace for memory operations."""
+    """Namespace for memory operations.
+    
+    Provides methods to query and retrieve stored memories.
+    """
 
     def __init__(self, client: Hindsight):
         self._client = client
@@ -171,8 +318,19 @@ class MemoriesAPI:
         search_query: str | None = None,
         limit: int = 100,
         offset: int = 0,
-    ):
-        """List memories in a bank."""
+    ) -> Any:
+        """List memories in a bank.
+        
+        Args:
+            bank_id: The ID of the bank to query.
+            type: Optional filter by memory type.
+            search_query: Optional search query for filtering.
+            limit: Maximum number of results to return (default: 100).
+            offset: Number of results to skip for pagination (default: 0).
+            
+        Returns:
+            List of memories matching the criteria.
+        """
         return self._client.list_memories(
             bank_id=bank_id,
             type=type,
@@ -205,9 +363,15 @@ class HindsightClient(Hindsight):
         directives = client.directives.list(bank_id="test")
         memories = client.memories.list(bank_id="test")
         ```
+        
+    Attributes:
+        banks: Namespace for bank management operations.
+        mental_models: Namespace for mental model operations.
+        directives: Namespace for directive operations.
+        memories: Namespace for memory listing operations.
     """
 
-    def __init__(self, *args, **kwargs):
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
         super().__init__(*args, **kwargs)
         self._banks_namespace: BanksAPI | None = None
         self._mental_models_namespace: MentalModelsAPI | None = None
@@ -216,28 +380,44 @@ class HindsightClient(Hindsight):
 
     @property
     def banks(self) -> BanksAPI:
-        """Access bank management operations."""
+        """Access bank management operations.
+        
+        Returns:
+            BanksAPI instance for bank operations.
+        """
         if self._banks_namespace is None:
             self._banks_namespace = BanksAPI(self)
         return self._banks_namespace
 
     @property
     def mental_models(self) -> MentalModelsAPI:
-        """Access mental model operations."""
+        """Access mental model operations.
+        
+        Returns:
+            MentalModelsAPI instance for mental model operations.
+        """
         if self._mental_models_namespace is None:
             self._mental_models_namespace = MentalModelsAPI(self)
         return self._mental_models_namespace
 
     @property
     def directives(self) -> DirectivesAPI:
-        """Access directive operations."""
+        """Access directive operations.
+        
+        Returns:
+            DirectivesAPI instance for directive operations.
+        """
         if self._directives_namespace is None:
             self._directives_namespace = DirectivesAPI(self)
         return self._directives_namespace
 
     @property
     def memories(self) -> MemoriesAPI:
-        """Access memory listing operations."""
+        """Access memory listing operations.
+        
+        Returns:
+            MemoriesAPI instance for memory operations.
+        """
         if self._memories_namespace is None:
             self._memories_namespace = MemoriesAPI(self)
         return self._memories_namespace

hindsight-all/hindsight/embedded.py

@@ -34,7 +34,6 @@ Using context manager:
 """
 
 import logging
-import os
 import threading
 from typing import Optional
 
@@ -74,6 +73,9 @@ class HindsightEmbedded:
         database_url: Optional database URL override (default: profile-specific pg0)
         idle_timeout: Seconds before daemon auto-exits when idle (default: 300)
         log_level: Daemon log level (default: "info")
+        ui: Whether to start the control plane web UI alongside the daemon (default: False)
+        ui_port: Port for the UI. Defaults to daemon_port + 10000.
+        ui_hostname: Hostname to bind the UI to. Defaults to "0.0.0.0".
     """
 
     def __init__(
@@ -86,6 +88,9 @@ class HindsightEmbedded:
         database_url: Optional[str] = None,
         idle_timeout: int = 300,
         log_level: str = "info",
+        ui: bool = False,
+        ui_port: Optional[int] = None,
+        ui_hostname: str = "0.0.0.0",
     ):
         """
         Initialize the embedded client (daemon starts on first use).
@@ -99,6 +104,9 @@ class HindsightEmbedded:
             database_url: Optional database URL override
             idle_timeout: Seconds before daemon auto-exits when idle
             log_level: Daemon log level
+            ui: Whether to start the control plane web UI alongside the daemon
+            ui_port: Port for the UI (defaults to daemon_port + 10000)
+            ui_hostname: Hostname to bind the UI to (defaults to "0.0.0.0")
         """
         self.profile = profile
 
@@ -117,6 +125,10 @@ class HindsightEmbedded:
         if database_url:
             self.config["HINDSIGHT_EMBED_API_DATABASE_URL"] = database_url
 
+        self._ui = ui
+        self._ui_port = ui_port
+        self._ui_hostname = ui_hostname
+
         self._client: Optional[Hindsight] = None
         self._lock = threading.Lock()
         self._started = False
@@ -140,13 +152,17 @@ class HindsightEmbedded:
                 return
 
             if self._closed:
-                raise RuntimeError("Cannot use HindsightEmbedded after it has been closed")
+                raise RuntimeError(
+                    "Cannot use HindsightEmbedded after it has been closed"
+                )
 
             # Use embed manager interface for daemon management
             logger.info(f"Ensuring daemon is running for profile '{self.profile}'...")
             success = self._manager.ensure_running(self.config, self.profile)
             if not success:
-                raise RuntimeError(f"Failed to start daemon for profile '{self.profile}'")
+                raise RuntimeError(
+                    f"Failed to start daemon for profile '{self.profile}'"
+                )
 
             # Get daemon URL and create client
             daemon_url = self._manager.get_url(self.profile)
@@ -154,6 +170,15 @@ class HindsightEmbedded:
             self._started = True
             logger.info(f"Connected to daemon at {daemon_url}")
 
+            # Start UI if requested
+            if self._ui:
+                logger.info(f"Starting UI for profile '{self.profile}'...")
+                ui_started = self._manager.start_ui(
+                    self.profile, self._ui_port, self._ui_hostname
+                )
+                if not ui_started:
+                    logger.warning(f"Failed to start UI for profile '{self.profile}'")
+
     def _cleanup(self, stop_daemon_on_close: bool = False):
         """
         Cleanup client resources (idempotent).
@@ -165,20 +190,47 @@ class HindsightEmbedded:
         if self._closed:
             return
 
-        with self._lock:
+        acquired = self._lock.acquire(timeout=5.0)
+        if not acquired:
+            # Lock is held by another thread (e.g. _ensure_started).
+            # Mark closed to prevent new operations but skip shared-state
+            # teardown — the daemon's idle timeout handles the rest.
+            logger.warning(
+                "Cleanup lock acquisition timed out for profile '%s'; "
+                "marking closed, daemon will idle-stop on its own",
+                self.profile,
+            )
+            self._closed = True
+            return
+
+        try:
             if self._closed:
                 return
 
             if self._client is not None:
-                self._client.close()
+                try:
+                    self._client.close()
+                except Exception:
+                    logger.debug(
+                        "Error closing client for profile '%s'",
+                        self.profile,
+                        exc_info=True,
+                    )
                 self._client = None
 
+            # Stop UI if it was started
+            if self._ui and self._started:
+                logger.info(f"Stopping UI for profile '{self.profile}'...")
+                self._manager.stop_ui(self.profile, self._ui_port)
+
             # Optionally stop daemon (daemon has idle timeout, so not required)
             if stop_daemon_on_close and self._started:
                 logger.info(f"Stopping daemon for profile '{self.profile}'...")
                 self._manager.stop(self.profile)
 
             self._closed = True
+        finally:
+            self._lock.release()
 
     def close(self, stop_daemon: bool = False):
         """
@@ -375,3 +427,8 @@ class HindsightEmbedded:
     def is_running(self) -> bool:
         """Check if the client is initialized."""
         return self._started and not self._closed and self._client is not None
+
+    @property
+    def ui_url(self) -> str:
+        """Get the UI URL for this profile."""
+        return self._manager.get_ui_url(self.profile)

hindsight-all/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "hindsight-all"
-version = "0.4.17"
+version = "0.5.1"
 description = "Hindsight: Agent Memory That Works Like Human Memory - All-in-One Bundle"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -20,6 +20,9 @@ hindsight-client = { workspace = true }
 hindsight-embed = { workspace = true }
 
 [project.optional-dependencies]
+local-llm = [
+    "hindsight-api-slim[local-llm]>=0.4.17",
+]
 test = [
     "pytest>=7.0.0",
     "pytest-asyncio>=0.21.0",

hindsight-all/tests/test_cleanup_timeout.py

@@ -0,0 +1,56 @@
+"""
+Unit test for _cleanup lock timeout behavior.
+
+Verifies that _cleanup completes even when the lock is held by another thread,
+instead of hanging indefinitely (fixes #952).
+"""
+
+import threading
+import time
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+
+def test_cleanup_completes_when_lock_held():
+    """
+    _cleanup should complete (best-effort) even when self._lock is held
+    by another thread, e.g. during a long _ensure_started call.
+    """
+    with patch.dict("sys.modules", {
+        "hindsight_client": MagicMock(),
+        "hindsight_embed": MagicMock(),
+        "hindsight.api_namespaces": MagicMock(),
+    }):
+        from hindsight.embedded import HindsightEmbedded
+
+        client = HindsightEmbedded.__new__(HindsightEmbedded)
+        client.profile = "test"
+        client._lock = threading.Lock()
+        client._closed = False
+        client._client = None
+        client._started = False
+        client._ui = False
+
+        # Simulate another thread holding the lock
+        client._lock.acquire()
+
+        cleanup_done = threading.Event()
+
+        def run_cleanup():
+            client._cleanup()
+            cleanup_done.set()
+
+        t = threading.Thread(target=run_cleanup)
+        t.start()
+
+        # Cleanup should complete within the timeout (5s) + margin
+        assert cleanup_done.wait(timeout=8.0), (
+            "_cleanup hung instead of timing out on lock acquisition"
+        )
+
+        # Release the lock from the simulating thread
+        client._lock.release()
+        t.join(timeout=1.0)
+
+        assert client._closed, "Client should be marked as closed after cleanup"

hindsight-all/tests/test_embedded.py

@@ -15,6 +15,8 @@ import os
 import uuid
 
 import pytest
+import urllib.request
+import json
 
 from hindsight import HindsightEmbedded
 
@@ -23,12 +25,20 @@ from hindsight import HindsightEmbedded
 def llm_config():
     """Get LLM configuration from environment (session-scoped)."""
     # Try both naming conventions
-    provider = os.getenv("HINDSIGHT_API_LLM_PROVIDER") or os.getenv("HINDSIGHT_LLM_PROVIDER", "groq")
-    api_key = os.getenv("HINDSIGHT_API_LLM_API_KEY") or os.getenv("HINDSIGHT_LLM_API_KEY", "")
-    model = os.getenv("HINDSIGHT_API_LLM_MODEL") or os.getenv("HINDSIGHT_LLM_MODEL", "openai/gpt-oss-120b")
+    provider = os.getenv("HINDSIGHT_API_LLM_PROVIDER") or os.getenv(
+        "HINDSIGHT_LLM_PROVIDER", "groq"
+    )
+    api_key = os.getenv("HINDSIGHT_API_LLM_API_KEY") or os.getenv(
+        "HINDSIGHT_LLM_API_KEY", ""
+    )
+    model = os.getenv("HINDSIGHT_API_LLM_MODEL") or os.getenv(
+        "HINDSIGHT_LLM_MODEL", "openai/gpt-oss-120b"
+    )
 
     if not api_key:
-        pytest.skip("LLM API key not configured. Set HINDSIGHT_API_LLM_API_KEY or HINDSIGHT_LLM_API_KEY.")
+        pytest.skip(
+            "LLM API key not configured. Set HINDSIGHT_API_LLM_API_KEY or HINDSIGHT_LLM_API_KEY."
+        )
 
     return {
         "llm_provider": provider,
@@ -78,7 +88,9 @@ def test_embedded_context_manager(llm_config):
 
         # Recall memory
         recall_results = client.recall(bank_id=bank_id, query="context")
-        assert isinstance(recall_results.results, list), "Recall should return results list"
+        assert isinstance(recall_results.results, list), (
+            "Recall should return results list"
+        )
 
     # Server should be stopped after context exit
     # Note: We can't check client.is_running here as client is out of scope
@@ -105,7 +117,9 @@ def test_embedded_complete_workflow(llm_config):
         # Step 1: Create a memory bank
         print(f"\n1. Creating memory bank: {bank_id}")
         bank_response = client.create_bank(
-            bank_id=bank_id, name="Test Assistant", mission="Help with programming tasks"
+            bank_id=bank_id,
+            name="Test Assistant",
+            mission="Help with programming tasks",
         )
         assert bank_response.bank_id == bank_id
 
@@ -126,7 +140,9 @@ def test_embedded_complete_workflow(llm_config):
             items=[
                 {"content": "User works with pandas and numpy."},
                 {"content": "User likes matplotlib for visualization."},
-                {"content": "User is interested in machine learning with scikit-learn."},
+                {
+                    "content": "User is interested in machine learning with scikit-learn."
+                },
             ],
         )
         assert batch_response.success
@@ -134,7 +150,9 @@ def test_embedded_complete_workflow(llm_config):
 
         # Step 4: Recall memories
         print("\n4. Recalling memories...")
-        recall_response = client.recall(bank_id=bank_id, query="What tools does the user prefer?", max_tokens=2000)
+        recall_response = client.recall(
+            bank_id=bank_id, query="What tools does the user prefer?", max_tokens=2000
+        )
         assert isinstance(recall_response.results, list)
         assert len(recall_response.results) > 0
         print(f"   Found {len(recall_response.results)} relevant memories")
@@ -152,7 +170,9 @@ def test_embedded_complete_workflow(llm_config):
 
         # Verify answer mentions relevant tools
         answer_lower = reflect_response.text.lower()
-        assert any(term in answer_lower for term in ["python", "pandas", "numpy", "data"])
+        assert any(
+            term in answer_lower for term in ["python", "pandas", "numpy", "data"]
+        )
 
         # Step 6: List memories
         print("\n6. Listing memories...")
@@ -215,7 +235,9 @@ def test_embedded_method_proxying(llm_config):
         assert bank.bank_id == bank_id
 
         # Test mission setting
-        mission_response = client.set_mission(bank_id=bank_id, mission="Test mission for proxying")
+        mission_response = client.set_mission(
+            bank_id=bank_id, mission="Test mission for proxying"
+        )
         assert mission_response.bank_id == bank_id
 
         # Test retain
@@ -264,7 +286,9 @@ def test_embedded_multiple_banks(llm_config):
 
         # Create second bank and store data
         client.create_bank(bank_id=bank2_id, name="Bank 2")
-        client.retain(bank_id=bank2_id, content="Bob uses JavaScript for web development")
+        client.retain(
+            bank_id=bank2_id, content="Bob uses JavaScript for web development"
+        )
 
         # Recall from both banks
         results1 = client.recall(bank_id=bank1_id, query="programming language")
@@ -275,9 +299,9 @@ def test_embedded_multiple_banks(llm_config):
 
         # Verify banks are isolated (each should only see their own content)
         # This is a basic check - content isolation is tested more thoroughly in other tests
-        assert results1.results[0].text != results2.results[0].text or len(results1.results) != len(
-            results2.results
-        )
+        assert results1.results[0].text != results2.results[0].text or len(
+            results1.results
+        ) != len(results2.results)
 
     finally:
         client.close()
@@ -296,10 +320,14 @@ def test_embedded_profile_isolation(llm_config):
 
     try:
         # Store data in profile1
-        client1.retain(bank_id=bank_id, content="User likes TypeScript for frontend development")
+        client1.retain(
+            bank_id=bank_id, content="User likes TypeScript for frontend development"
+        )
 
         # Store different data in profile2
-        client2.retain(bank_id=bank_id, content="User prefers Rust for systems programming")
+        client2.retain(
+            bank_id=bank_id, content="User prefers Rust for systems programming"
+        )
 
         # Each profile should only see its own data
         results1 = client1.recall(bank_id=bank_id, query="programming preference")
@@ -334,5 +362,42 @@ def test_embedded_error_after_close(llm_config):
     assert not client.is_running
 
     # Trying to use it after close should raise an error
-    with pytest.raises(RuntimeError, match="Cannot use HindsightEmbedded after it has been closed"):
+    with pytest.raises(
+        RuntimeError, match="Cannot use HindsightEmbedded after it has been closed"
+    ):
         client.retain(bank_id=bank_id, content="This should fail")
+
+
+def test_embedded_ui_flag(llm_config):
+    """
+    Test that ui=True starts the control plane UI alongside the daemon,
+    and that the UI's health endpoint reports a connected dataplane.
+    """
+    profile = f"test_ui_{uuid.uuid4().hex[:8]}"
+    bank_id = f"bank_{uuid.uuid4().hex[:8]}"
+
+    client = HindsightEmbedded(profile=profile, log_level="info", ui=True, **llm_config)
+
+    try:
+        # First use triggers daemon + UI startup
+        result = client.retain(bank_id=bank_id, content="UI integration test content")
+        assert result.success, "Retain should succeed"
+        assert client.is_running, "Daemon should be running"
+
+        # Verify UI is reachable and reports connected dataplane
+        ui_url = client.ui_url
+        assert ui_url, "ui_url should be set"
+
+        health_url = f"{ui_url}/api/health"
+        with urllib.request.urlopen(health_url, timeout=10) as resp:
+            health = json.loads(resp.read().decode())
+
+        assert health["status"] == "ok", (
+            f"UI health status should be 'ok', got: {health['status']}"
+        )
+        assert health["dataplane"]["status"] == "connected", (
+            f"Dataplane should be connected, got: {health['dataplane']}"
+        )
+
+    finally:
+        client.close()

hindsight-api-slim/hindsight_api/__init__.py

@@ -46,4 +46,4 @@ __all__ = [
     "RemoteTEICrossEncoder",
     "LLMConfig",
 ]
-__version__ = "0.4.17"
+__version__ = "0.5.1"

hindsight-api-slim/hindsight_api/admin/cli.py

@@ -249,7 +249,7 @@ async def _run_migration(
         schemas = list(dict.fromkeys(schemas))
 
     for schema in schemas:
-        run_migrations(resolved_url, schema=schema)
+        run_migrations(resolved_url, schema=schema, migration_database_url=config.migration_database_url)
 
     if embedding_dimension is not None:
         for schema in schemas:

hindsight-api-slim/hindsight_api/alembic/versions/2eee35aa3cfc_case_insensitive_entities_trgm_index.py

@@ -0,0 +1,45 @@
+"""Recreate entities trigram index on LOWER(canonical_name) for case-insensitive matching
+
+The previous GIN trigram index on canonical_name was case-sensitive, causing
+"Alice" and "alice" to have different trigram sets. This recreates it on
+LOWER(canonical_name) so the % operator matches case-insensitively.
+
+Revision ID: d6e7f8a9b0c1
+Revises: c5d6e7f8a9b0
+Create Date: 2026-03-31
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "d6e7f8a9b0c1"
+down_revision: str | Sequence[str] | None = "c5d6e7f8a9b0"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+    # Drop the old case-sensitive trigram index
+    op.execute("DROP INDEX IF EXISTS entities_canonical_name_trgm_idx")
+    # Create case-insensitive trigram index on LOWER(canonical_name)
+    op.execute(
+        f"CREATE INDEX IF NOT EXISTS entities_canonical_name_lower_trgm_idx "
+        f"ON {schema}entities USING GIN (LOWER(canonical_name) gin_trgm_ops)"
+    )
+
+
+def downgrade() -> None:
+    op.execute("DROP INDEX IF EXISTS entities_canonical_name_lower_trgm_idx")
+    schema = _get_schema_prefix()
+    # Restore original case-sensitive index
+    op.execute(
+        f"CREATE INDEX IF NOT EXISTS entities_canonical_name_trgm_idx "
+        f"ON {schema}entities USING GIN (canonical_name gin_trgm_ops)"
+    )

hindsight-api-slim/hindsight_api/alembic/versions/a3b4c5d6e7f8_add_consolidation_failed_at_to_memory_units.py

@@ -0,0 +1,52 @@
+"""Add consolidation_failed_at column to memory_units for tracking persistent LLM failures.
+
+When all LLM retries are exhausted on a single-memory batch, the memory is marked
+with consolidation_failed_at instead of consolidated_at, so it is not silently lost
+and can be retried later via the API.
+
+Revision ID: a3b4c5d6e7f8
+Revises: g7h8i9j0k1l2
+Create Date: 2026-03-17
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "a3b4c5d6e7f8"
+down_revision: str | Sequence[str] | None = "g7h8i9j0k1l2"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    """Get schema prefix for table names (required for multi-tenant support)."""
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+
+    op.execute(
+        f"""
+        ALTER TABLE {schema}memory_units
+        ADD COLUMN IF NOT EXISTS consolidation_failed_at TIMESTAMPTZ DEFAULT NULL
+        """
+    )
+
+    # Index to efficiently query memories that failed consolidation for a given bank
+    op.execute(
+        f"""
+        CREATE INDEX IF NOT EXISTS idx_memory_units_consolidation_failed
+        ON {schema}memory_units (bank_id, consolidation_failed_at)
+        WHERE consolidation_failed_at IS NOT NULL AND fact_type IN ('experience', 'world')
+        """
+    )
+
+
+def downgrade() -> None:
+    schema = _get_schema_prefix()
+
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_memory_units_consolidation_failed")
+    op.execute(f"ALTER TABLE {schema}memory_units DROP COLUMN IF EXISTS consolidation_failed_at")

hindsight-api-slim/hindsight_api/alembic/versions/a4b5c6d7e8f9_fix_per_bank_vector_index_type.py

@@ -0,0 +1,142 @@
+"""Fix per-bank vector indexes to match configured extension
+
+Revision ID: a4b5c6d7e8f9
+Revises: d6e7f8a9b0c1
+Create Date: 2026-04-01
+
+Migration d5e6f7a8b9c0 hardcoded HNSW when creating per-bank partial vector
+indexes, ignoring HINDSIGHT_API_VECTOR_EXTENSION. Banks that existed when that
+migration ran got HNSW indexes even when pgvectorscale (DiskANN) or vchord
+was configured.
+
+This migration detects the mismatch and recreates the affected indexes with
+the correct type. Skipped entirely when the configured extension is pgvector
+(the default), since those indexes are already correct.
+"""
+
+import os
+from collections.abc import Sequence
+
+from alembic import context, op
+from sqlalchemy import text
+
+revision: str = "a4b5c6d7e8f9"
+down_revision: str | Sequence[str] | None = "d6e7f8a9b0c1"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_FACT_TYPES: dict[str, str] = {
+    "world": "worl",
+    "experience": "expr",
+    "observation": "obsv",
+}
+
+
+def _get_schema_prefix() -> str:
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def _target_index_type() -> str | None:
+    """Return the target index type, or None if pgvector (no fix needed)."""
+    ext = os.getenv("HINDSIGHT_API_VECTOR_EXTENSION", "pgvector").lower()
+    if ext == "pgvectorscale":
+        return "diskann"
+    elif ext == "vchord":
+        return "vchordrq"
+    return None
+
+
+def _vector_index_using_clause() -> str:
+    """Return the USING clause based on the configured vector extension."""
+    ext = os.getenv("HINDSIGHT_API_VECTOR_EXTENSION", "pgvector").lower()
+    if ext == "pgvectorscale":
+        return "USING diskann (embedding vector_cosine_ops) WITH (num_neighbors = 50)"
+    elif ext == "vchord":
+        return "USING vchordrq (embedding vector_l2_ops)"
+    else:
+        return "USING hnsw (embedding vector_cosine_ops)"
+
+
+def upgrade() -> None:
+    target = _target_index_type()
+    if target is None:
+        # pgvector — indexes are already HNSW, nothing to fix
+        return
+
+    bind = op.get_bind()
+    schema_name = context.config.get_main_option("target_schema")
+    schema = _get_schema_prefix()
+    table_ref = f'"{schema_name}".memory_units' if schema_name else "memory_units"
+    banks_ref = f'"{schema_name}".banks' if schema_name else "banks"
+    using_clause = _vector_index_using_clause()
+    pg_schema = schema_name or "public"
+
+    rows = bind.execute(text(f"SELECT bank_id, internal_id FROM {banks_ref}")).fetchall()  # noqa: S608
+    for row in rows:
+        bank_id = row[0]
+        internal_id = str(row[1]).replace("-", "")[:16]
+        escaped_bank_id = bank_id.replace("'", "''")
+        for ft, ft_short in _FACT_TYPES.items():
+            idx_name = f"idx_mu_emb_{ft_short}_{internal_id}"
+
+            # Check if this index exists and what type it is
+            idx_info = bind.execute(
+                text("SELECT indexdef FROM pg_indexes WHERE schemaname = :schema AND indexname = :idx"),
+                {"schema": pg_schema, "idx": idx_name},
+            ).fetchone()
+
+            if idx_info is None:
+                # Index doesn't exist — create it with the correct type
+                bind.execute(
+                    text(
+                        f"CREATE INDEX IF NOT EXISTS {idx_name} "
+                        f"ON {table_ref} {using_clause} "
+                        f"WHERE fact_type = '{ft}' AND bank_id = '{escaped_bank_id}'"
+                    )
+                )
+                continue
+
+            indexdef = idx_info[0].lower()
+            if target in indexdef:
+                # Already the correct type
+                continue
+
+            # Wrong type — drop and recreate
+            bind.execute(text(f"DROP INDEX IF EXISTS {schema}{idx_name}"))
+            bind.execute(
+                text(
+                    f"CREATE INDEX IF NOT EXISTS {idx_name} "
+                    f"ON {table_ref} {using_clause} "
+                    f"WHERE fact_type = '{ft}' AND bank_id = '{escaped_bank_id}'"
+                )
+            )
+
+
+def downgrade() -> None:
+    # Downgrade recreates indexes as HNSW (the original hardcoded behavior)
+    target = _target_index_type()
+    if target is None:
+        return
+
+    bind = op.get_bind()
+    schema_name = context.config.get_main_option("target_schema")
+    schema = _get_schema_prefix()
+    table_ref = f'"{schema_name}".memory_units' if schema_name else "memory_units"
+    banks_ref = f'"{schema_name}".banks' if schema_name else "banks"
+
+    rows = bind.execute(text(f"SELECT bank_id, internal_id FROM {banks_ref}")).fetchall()  # noqa: S608
+    for row in rows:
+        bank_id = row[0]
+        internal_id = str(row[1]).replace("-", "")[:16]
+        escaped_bank_id = bank_id.replace("'", "''")
+        for ft, ft_short in _FACT_TYPES.items():
+            idx_name = f"idx_mu_emb_{ft_short}_{internal_id}"
+            bind.execute(text(f"DROP INDEX IF EXISTS {schema}{idx_name}"))
+            bind.execute(
+                text(
+                    f"CREATE INDEX IF NOT EXISTS {idx_name} "
+                    f"ON {table_ref} USING hnsw (embedding vector_cosine_ops) "
+                    f"WHERE fact_type = '{ft}' AND bank_id = '{escaped_bank_id}'"
+                )
+            )

hindsight-api-slim/hindsight_api/alembic/versions/b3c4d5e6f7a8_add_content_hash_to_chunks.py

@@ -0,0 +1,32 @@
+"""add content_hash to chunks table for delta retain
+
+Revision ID: b3c4d5e6f7a8
+Revises: a3b4c5d6e7f8
+Create Date: 2026-03-25
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "b3c4d5e6f7a8"
+down_revision: str | Sequence[str] | None = "a3b4c5d6e7f8"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    """Get schema prefix for table names (required for multi-tenant support)."""
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+    # Add content_hash column to chunks table for delta comparison
+    op.execute(f"ALTER TABLE {schema}chunks ADD COLUMN IF NOT EXISTS content_hash TEXT")
+
+
+def downgrade() -> None:
+    schema = _get_schema_prefix()
+    op.execute(f"ALTER TABLE {schema}chunks DROP COLUMN IF EXISTS content_hash")

hindsight-api-slim/hindsight_api/alembic/versions/c1a2b3d4e5f6_enable_pg_trgm_and_entities_trgm_index.py

@@ -11,6 +11,7 @@ block; see migrations.py for how this is handled safely.
 
 from collections.abc import Sequence
 
+import sqlalchemy as sa
 from alembic import context, op
 
 revision: str = "c1a2b3d4e5f6"
@@ -25,9 +26,21 @@ def _get_schema_prefix() -> str:
 
 
 def upgrade() -> None:
-    # pg_trgm ships with every standard PostgreSQL installation as a contrib module.
+    # pg_trgm ships with most PostgreSQL installations as a contrib module.
     # It enables fast similarity lookups via GIN indexes, used for entity name matching.
-    op.execute("CREATE EXTENSION IF NOT EXISTS pg_trgm")
+    # On managed services (e.g. Azure Flexible Server), the extension may not be
+    # available or may require manual enablement.  We gracefully skip the index
+    # creation if the extension cannot be loaded — the entity resolver will
+    # auto-detect and fall back to the "full" lookup strategy at runtime.  See #626.
+    conn = op.get_bind()
+    try:
+        conn.execute(sa.text("CREATE EXTENSION IF NOT EXISTS pg_trgm"))
+    except Exception:
+        # Extension not available (managed Postgres, insufficient privileges, etc.)
+        # Roll back the failed statement and skip index creation.
+        conn.execute(sa.text("ROLLBACK"))
+        conn.execute(sa.text("BEGIN"))
+        return
 
     schema = _get_schema_prefix()
     # GIN index on canonical_name enables sub-millisecond trigram similarity queries

hindsight-api-slim/hindsight_api/alembic/versions/c2d3e4f5g6h7_add_audit_log_table.py

@@ -0,0 +1,61 @@
+"""Add audit_log table for feature usage tracking.
+
+Merge migration that combines the two existing heads (a3b4c5d6e7f8 + c8e5f2a3b4d1).
+
+Stores raw request/response as JSONB for expandability without future migrations.
+The metadata JSONB column allows adding arbitrary fields in the future.
+
+Revision ID: c2d3e4f5g6h7
+Revises: a3b4c5d6e7f8, c8e5f2a3b4d1
+Create Date: 2026-03-26
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "c2d3e4f5g6h7"
+down_revision: str | Sequence[str] | None = ("a3b4c5d6e7f8", "c8e5f2a3b4d1")
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    """Get schema prefix for table names (required for multi-tenant support)."""
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+
+    op.execute(
+        f"""
+        CREATE TABLE IF NOT EXISTS {schema}audit_log (
+            id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
+            action TEXT NOT NULL,
+            transport TEXT NOT NULL,
+            bank_id TEXT,
+            started_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+            ended_at TIMESTAMPTZ,
+            request JSONB,
+            response JSONB,
+            metadata JSONB DEFAULT '{{}}'::jsonb
+        )
+        """
+    )
+
+    op.execute(
+        f"CREATE INDEX IF NOT EXISTS idx_audit_log_action_started ON {schema}audit_log (action, started_at DESC)"
+    )
+    op.execute(f"CREATE INDEX IF NOT EXISTS idx_audit_log_bank_started ON {schema}audit_log (bank_id, started_at DESC)")
+    op.execute(f"CREATE INDEX IF NOT EXISTS idx_audit_log_started ON {schema}audit_log (started_at DESC)")
+
+
+def downgrade() -> None:
+    schema = _get_schema_prefix()
+
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_audit_log_started")
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_audit_log_bank_started")
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_audit_log_action_started")
+    op.execute(f"DROP TABLE IF EXISTS {schema}audit_log")

hindsight-api-slim/hindsight_api/alembic/versions/c5d6e7f8a9b0_add_bank_id_to_memory_links.py

@@ -0,0 +1,48 @@
+"""Add bank_id column to memory_links for direct filtering
+
+The stats endpoint JOINs memory_links to memory_units just to filter by
+bank_id.  With millions of links this takes 18+ seconds.  Adding bank_id
+directly to memory_links lets Postgres push the filter down before the JOIN.
+
+Revision ID: c5d6e7f8a9b0
+Revises: b3c4d5e6f7a8
+Create Date: 2026-03-26
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "c5d6e7f8a9b0"
+down_revision: str | Sequence[str] | None = "b3c4d5e6f7a8"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+
+    # 1. Add nullable column
+    op.execute(f"ALTER TABLE {schema}memory_links ADD COLUMN IF NOT EXISTS bank_id TEXT")
+
+    # 2. Backfill from memory_units
+    op.execute(f"""
+        UPDATE {schema}memory_links ml
+        SET bank_id = mu.bank_id
+        FROM {schema}memory_units mu
+        WHERE ml.from_unit_id = mu.id
+          AND ml.bank_id IS NULL
+    """)
+
+    # 3. Set NOT NULL
+    op.execute(f"ALTER TABLE {schema}memory_links ALTER COLUMN bank_id SET NOT NULL")
+
+
+def downgrade() -> None:
+    schema = _get_schema_prefix()
+    op.execute(f"ALTER TABLE {schema}memory_links DROP COLUMN IF EXISTS bank_id")

hindsight-api-slim/hindsight_api/alembic/versions/d5e6f7a8b9c0_add_bank_internal_id_and_per_bank_hnsw.py

@@ -1,4 +1,4 @@
-"""Add internal_id to banks and per-(bank, fact_type) partial HNSW indexes
+"""Add internal_id to banks and per-(bank, fact_type) partial vector indexes
 
 Revision ID: d5e6f7a8b9c0
 Revises: a3b4c5d6e7f8
@@ -6,25 +6,20 @@ Create Date: 2026-03-11
 
 This migration:
 1. Adds internal_id UUID column to banks (stable identifier for index naming)
-2. Drops the global HNSW index (competes with per-bank partial indexes)
-3. Creates per-(bank_id, fact_type) partial HNSW indexes for all existing banks
-   (new banks get indexes created at bank-creation time via bank_utils.create_bank_hnsw_indexes)
+2. Drops the global vector index (competes with per-bank partial indexes)
+3. Creates per-(bank_id, fact_type) partial vector indexes for all existing banks
+   using the configured vector extension (HNSW for pgvector, DiskANN for
+   pgvectorscale, vchordrq for vchord).
+   (new banks get indexes created at bank-creation time via bank_utils.create_bank_vector_indexes)
 
 Why per-(bank, fact_type) indexes:
 - fact_type-only partial indexes are never chosen by the planner when bank_id is in the WHERE
   clause, because the idx_memory_units_bank_id B-tree index always wins at planning time.
 - Per-(bank, fact_type) partial indexes have both predicates matching → planner selects them.
-- The global HNSW index competes for larger partitions (world, observation) and must be dropped.
-
-For large deployments, create indexes CONCURRENTLY before running this migration:
-    SELECT internal_id, bank_id FROM banks;
-    -- for each bank and each fact_type in (world, experience, observation):
-    CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_mu_emb_{ft}_{uid16}
-        ON memory_units USING hnsw (embedding vector_cosine_ops)
-        WHERE fact_type = '{ft}' AND bank_id = '{bank_id}';
-    DROP INDEX CONCURRENTLY IF EXISTS idx_memory_units_embedding;
+- The global vector index competes for larger partitions (world, observation) and must be dropped.
 """
 
+import os
 from collections.abc import Sequence
 
 from alembic import context, op
@@ -35,7 +30,7 @@ down_revision: str | Sequence[str] | None = "c3d4e5f6g7h8"
 branch_labels: str | Sequence[str] | None = None
 depends_on: str | Sequence[str] | None = None
 
-_HNSW_FACT_TYPES: dict[str, str] = {
+_FACT_TYPES: dict[str, str] = {
     "world": "worl",
     "experience": "expr",
     "observation": "obsv",
@@ -47,6 +42,17 @@ def _get_schema_prefix() -> str:
     return f'"{schema}".' if schema else ""
 
 
+def _vector_index_using_clause() -> str:
+    """Return the USING clause based on the configured vector extension."""
+    ext = os.getenv("HINDSIGHT_API_VECTOR_EXTENSION", "pgvector").lower()
+    if ext == "pgvectorscale":
+        return "USING diskann (embedding vector_cosine_ops) WITH (num_neighbors = 50)"
+    elif ext == "vchord":
+        return "USING vchordrq (embedding vector_l2_ops)"
+    else:
+        return "USING hnsw (embedding vector_cosine_ops)"
+
+
 def upgrade() -> None:
     schema = _get_schema_prefix()
 
@@ -56,33 +62,35 @@ def upgrade() -> None:
     )
     op.execute(f"ALTER TABLE {schema}banks ADD CONSTRAINT banks_internal_id_unique UNIQUE (internal_id)")
 
-    # 2. Drop any fact_type-only partial HNSW indexes that may exist from prior migrations
+    # 2. Drop any fact_type-only partial indexes that may exist from prior migrations
     #    (bank_id B-tree always wins over them when bank_id is in the WHERE clause)
     op.execute(f"DROP INDEX IF EXISTS {schema}idx_mu_emb_world")
     op.execute(f"DROP INDEX IF EXISTS {schema}idx_mu_emb_observation")
     op.execute(f"DROP INDEX IF EXISTS {schema}idx_mu_emb_experience")
 
-    # 4. Drop global HNSW index (competes with per-bank partial indexes)
+    # 4. Drop global vector index (competes with per-bank partial indexes)
     op.execute(f"DROP INDEX IF EXISTS {schema}idx_memory_units_embedding")
 
-    # 5. Create per-(bank, fact_type) partial HNSW indexes for all existing banks
+    # 5. Create per-(bank, fact_type) partial vector indexes for all existing banks
+    #    using the configured extension (HNSW / DiskANN / vchordrq)
     bind = op.get_bind()
     schema_name = context.config.get_main_option("target_schema")
     table_ref = f'"{schema_name}".memory_units' if schema_name else "memory_units"
     banks_ref = f'"{schema_name}".banks' if schema_name else "banks"
+    using_clause = _vector_index_using_clause()
 
     rows = bind.execute(text(f"SELECT bank_id, internal_id FROM {banks_ref}")).fetchall()  # noqa: S608
     for row in rows:
         bank_id = row[0]
         internal_id = str(row[1]).replace("-", "")[:16]
         escaped_bank_id = bank_id.replace("'", "''")
-        for ft, ft_short in _HNSW_FACT_TYPES.items():
+        for ft, ft_short in _FACT_TYPES.items():
             idx_name = f"idx_mu_emb_{ft_short}_{internal_id}"
             # Index name is schema-unqualified (indexes live in the schema of their table)
             bind.execute(
                 text(
                     f"CREATE INDEX IF NOT EXISTS {idx_name} "
-                    f"ON {table_ref} USING hnsw (embedding vector_cosine_ops) "
+                    f"ON {table_ref} {using_clause} "
                     f"WHERE fact_type = '{ft}' AND bank_id = '{escaped_bank_id}'"
                 )
             )

hindsight-api-slim/hindsight_api/alembic/versions/f1a2b3c4d5e6_add_memory_links_composite_index.py

@@ -5,7 +5,7 @@ Revises: e0a1b2c3d4e5
 Create Date: 2025-01-12
 
 Add composite index on memory_links (from_unit_id, link_type, weight DESC)
-to optimize MPFP graph traversal queries that need top-k edges per type.
+to optimize graph traversal queries that need top-k edges per type.
 """
 
 from collections.abc import Sequence
@@ -26,7 +26,7 @@ def _get_schema_prefix() -> str:
 
 
 def upgrade() -> None:
-    """Add composite index for efficient MPFP edge loading."""
+    """Add composite index for efficient graph retrieval edge loading."""
     schema = _get_schema_prefix()
     # Create composite index for efficient top-k per (from_node, link_type) queries
     # This enables LATERAL joins to use index-only scans with early termination

hindsight-api-slim/hindsight_api/alembic/versions/f6g7h8i9j0k1_chunk_fk_cascade_delete.py

@@ -0,0 +1,57 @@
+"""chunk_fk_cascade_delete
+
+Revision ID: f6g7h8i9j0k1
+Revises: e5f6g7h8i9j0
+Create Date: 2026-03-16 00:00:00.000000
+
+"""
+
+from collections.abc import Sequence
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "f6g7h8i9j0k1"
+down_revision: str | Sequence[str] | None = "e5f6g7h8i9j0"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    """Change memory_units.chunk_id FK from SET NULL to CASCADE.
+
+    When a document is deleted the CASCADE reaches chunks first; with SET NULL
+    the memory_units rows survived with chunk_id = NULL, leaving ghost records.
+    Switching to CASCADE ensures they are removed together with their chunk.
+    """
+    from alembic import context
+
+    schema = context.config.get_main_option("target_schema")
+    schema_prefix = f'"{schema}".' if schema else ""
+    # Use raw SQL with IF EXISTS so this is safe on schemas where the FK was
+    # already dropped or never existed under this name.
+    op.execute(f"ALTER TABLE {schema_prefix}memory_units DROP CONSTRAINT IF EXISTS memory_units_chunk_fkey")
+    # Use a DO block so the ADD is also idempotent: if the FK already exists (e.g.
+    # the schema was provisioned after the base migration already added it) the
+    # duplicate_object exception is swallowed rather than failing the migration.
+    op.execute(
+        f"""
+        DO $$ BEGIN
+            ALTER TABLE {schema_prefix}memory_units
+                ADD CONSTRAINT memory_units_chunk_fkey
+                FOREIGN KEY (chunk_id)
+                REFERENCES {schema_prefix}chunks (chunk_id)
+                ON DELETE CASCADE;
+        EXCEPTION
+            WHEN duplicate_object THEN NULL;
+        END $$;
+        """
+    )
+
+
+def downgrade() -> None:
+    """Revert to SET NULL behaviour."""
+    op.drop_constraint("memory_units_chunk_fkey", "memory_units", type_="foreignkey")
+    op.create_foreign_key(
+        "memory_units_chunk_fkey", "memory_units", "chunks", ["chunk_id"], ["chunk_id"], ondelete="SET NULL"
+    )

hindsight-api-slim/hindsight_api/alembic/versions/g2h3i4j5k6l7_remove_opinion_fact_type.py

@@ -0,0 +1,83 @@
+"""remove_opinion_fact_type
+
+Revision ID: g2h3i4j5k6l7
+Revises: f1a2b3c4d5e6
+Create Date: 2026-04-02
+
+Remove the deprecated 'opinion' fact type: drop opinion-specific indexes,
+update CHECK constraints, delete any remaining opinion rows, and drop the
+confidence_score column (was only used for opinions, always NULL otherwise).
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+# revision identifiers, used by Alembic.
+revision: str = "g2h3i4j5k6l7"
+down_revision: str | Sequence[str] | None = "f1a2b3c4d5e6"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    """Get schema prefix for table names (required for multi-tenant support)."""
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+
+    # 1. Delete any remaining opinion rows
+    op.execute(f"DELETE FROM {schema}memory_units WHERE fact_type = 'opinion'")
+
+    # 2. Drop opinion-specific indexes
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_memory_units_opinion_confidence")
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_memory_units_opinion_date")
+
+    # 3. Drop confidence_score constraints and column (only used for opinions, always NULL otherwise)
+    op.execute(f"ALTER TABLE {schema}memory_units DROP CONSTRAINT IF EXISTS confidence_score_fact_type_check")
+    op.execute(f"ALTER TABLE {schema}memory_units DROP CONSTRAINT IF EXISTS memory_units_confidence_score_check")
+    op.execute(f"ALTER TABLE {schema}memory_units DROP COLUMN IF EXISTS confidence_score")
+
+    # 4. Replace fact_type CHECK constraint
+    op.execute(f"ALTER TABLE {schema}memory_units DROP CONSTRAINT IF EXISTS memory_units_fact_type_check")
+    op.execute(
+        f"ALTER TABLE {schema}memory_units ADD CONSTRAINT memory_units_fact_type_check "
+        f"CHECK (fact_type IN ('world', 'experience', 'observation'))"
+    )
+
+
+def downgrade() -> None:
+    schema = _get_schema_prefix()
+
+    # Restore confidence_score column
+    op.execute(f"ALTER TABLE {schema}memory_units ADD COLUMN IF NOT EXISTS confidence_score float")
+    op.execute(
+        f"ALTER TABLE {schema}memory_units ADD CONSTRAINT memory_units_confidence_score_check "
+        f"CHECK (confidence_score IS NULL OR (confidence_score >= 0.0 AND confidence_score <= 1.0))"
+    )
+    op.execute(
+        f"ALTER TABLE {schema}memory_units ADD CONSTRAINT confidence_score_fact_type_check "
+        f"CHECK ((fact_type = 'opinion' AND confidence_score IS NOT NULL) OR "
+        f"(fact_type = 'observation') OR "
+        f"(fact_type NOT IN ('opinion', 'observation') AND confidence_score IS NULL))"
+    )
+
+    # Restore original fact_type CHECK constraint (with opinion)
+    op.execute(f"ALTER TABLE {schema}memory_units DROP CONSTRAINT IF EXISTS memory_units_fact_type_check")
+    op.execute(
+        f"ALTER TABLE {schema}memory_units ADD CONSTRAINT memory_units_fact_type_check "
+        f"CHECK (fact_type IN ('world', 'experience', 'opinion', 'observation'))"
+    )
+
+    # Recreate opinion indexes
+    op.execute(
+        f"CREATE INDEX idx_memory_units_opinion_confidence ON {schema}memory_units "
+        f"(bank_id, confidence_score DESC) WHERE fact_type = 'opinion'"
+    )
+    op.execute(
+        f"CREATE INDEX idx_memory_units_opinion_date ON {schema}memory_units "
+        f"(bank_id, event_date DESC) WHERE fact_type = 'opinion'"
+    )

hindsight-api-slim/hindsight_api/alembic/versions/g7h8i9j0k1l2_backsweep_orphan_observations.py

@@ -0,0 +1,71 @@
+"""backsweep_orphan_memory_units
+
+Two-pass cleanup of memory_units rows that were never removed by earlier bugs:
+
+Pass 1 — any fact_type, bank gone:
+  memory_units whose bank_id no longer exists in banks.  These accumulate when
+  a bank is deleted without a proper cascade (no FK from memory_units to banks
+  exists in the schema).
+
+Pass 2 — observations only, all sources gone:
+  observation rows whose bank still exists but every source_memory_id points
+  to a deleted memory unit.  These were left behind before PR #580 fixed the
+  chunk FK cascade and before delete_document() called
+  _delete_stale_observations_for_memories.
+
+Revision ID: g7h8i9j0k1l2
+Revises: f6g7h8i9j0k1
+Create Date: 2026-03-16
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "g7h8i9j0k1l2"
+down_revision: str | Sequence[str] | None = "f6g7h8i9j0k1"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+    mu = f"{schema}memory_units"
+    banks = f"{schema}banks"
+
+    # Pass 1: delete all memory_units (any fact_type) whose bank no longer exists.
+    # There is no FK from memory_units to banks, so these never cascade away.
+    op.execute(
+        f"""
+        DELETE FROM {mu}
+        WHERE NOT EXISTS (
+            SELECT 1 FROM {banks} b WHERE b.bank_id = {mu}.bank_id
+        )
+        """
+    )
+
+    # Pass 2: delete orphaned observations whose bank still exists but every
+    # source_memory_id refers to a now-deleted memory unit (or the array is
+    # empty).  Observations with at least one surviving source are left alone.
+    op.execute(
+        f"""
+        DELETE FROM {mu} orphan
+        WHERE orphan.fact_type = 'observation'
+          AND NOT EXISTS (
+              SELECT 1
+              FROM {mu} src
+              WHERE src.id = ANY(orphan.source_memory_ids)
+                AND src.bank_id = orphan.bank_id
+          )
+        """
+    )
+
+
+def downgrade() -> None:
+    # Deleted rows cannot be restored.
+    pass

hindsight-api-slim/hindsight_api/alembic/versions/h3i4j5k6l7m8_merge_heads_and_add_unit_entities_index.py

@@ -0,0 +1,42 @@
+"""Merge 3 migration heads and add unit_entities composite index
+
+Revision ID: h3i4j5k6l7m8
+Revises: a4b5c6d7e8f9, c2d3e4f5g6h7, g2h3i4j5k6l7
+Create Date: 2026-04-07
+
+Merges three unmerged migration heads into one, and adds a composite index
+(entity_id, unit_id) on unit_entities for index-only scans in the LATERAL
+entity expansion query.
+"""
+
+from collections.abc import Sequence
+
+from alembic import context, op
+
+revision: str = "h3i4j5k6l7m8"
+down_revision: str | Sequence[str] | None = ("a4b5c6d7e8f9", "c2d3e4f5g6h7", "g2h3i4j5k6l7")
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def _get_schema_prefix() -> str:
+    """Get schema prefix for table names (required for multi-tenant support)."""
+    schema = context.config.get_main_option("target_schema")
+    return f'"{schema}".' if schema else ""
+
+
+def upgrade() -> None:
+    schema = _get_schema_prefix()
+    # Composite index enables index-only scans for entity_id -> unit_id lookups
+    op.execute(
+        f"CREATE INDEX IF NOT EXISTS idx_unit_entities_entity_unit ON {schema}unit_entities (entity_id, unit_id)"
+    )
+    # Drop the now-redundant single-column index
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_unit_entities_entity")
+
+
+def downgrade() -> None:
+    schema = _get_schema_prefix()
+    op.execute(f"DROP INDEX IF EXISTS {schema}idx_unit_entities_entity_unit")
+    # Restore the single-column index
+    op.execute(f"CREATE INDEX IF NOT EXISTS idx_unit_entities_entity ON {schema}unit_entities (entity_id)")

hindsight-api-slim/hindsight_api/api/mcp.py

@@ -12,44 +12,9 @@ from hindsight_api.config import _get_raw_config
 from hindsight_api.engine.memory_engine import _current_schema
 from hindsight_api.extensions import MCPExtension, load_extension
 from hindsight_api.extensions.tenant import AuthenticationError
-from hindsight_api.mcp_tools import MCPToolsConfig, register_mcp_tools
+from hindsight_api.mcp_tools import _ALL_TOOLS, MCPToolsConfig, register_mcp_tools
 from hindsight_api.models import RequestContext
 
-# All tools available in the system (explicit list — no wildcards)
-_ALL_TOOLS: frozenset[str] = frozenset(
-    {
-        "retain",
-        "recall",
-        "reflect",
-        "list_banks",
-        "create_bank",
-        "list_mental_models",
-        "get_mental_model",
-        "create_mental_model",
-        "update_mental_model",
-        "delete_mental_model",
-        "refresh_mental_model",
-        "list_directives",
-        "create_directive",
-        "delete_directive",
-        "list_memories",
-        "get_memory",
-        "delete_memory",
-        "list_documents",
-        "get_document",
-        "delete_document",
-        "list_operations",
-        "get_operation",
-        "cancel_operation",
-        "list_tags",
-        "get_bank",
-        "get_bank_stats",
-        "update_bank",
-        "delete_bank",
-        "clear_memories",
-    }
-)
-
 # Configure logging from HINDSIGHT_API_LOG_LEVEL environment variable
 _log_level_str = os.environ.get("HINDSIGHT_API_LOG_LEVEL", "info").lower()
 _log_level_map = {
@@ -83,6 +48,9 @@ _current_api_key: ContextVar[str | None] = ContextVar("current_api_key", default
 _current_tenant_id: ContextVar[str | None] = ContextVar("current_tenant_id", default=None)
 _current_api_key_id: ContextVar[str | None] = ContextVar("current_api_key_id", default=None)
 
+# Context variable for MCP pre-authentication flag (set when MCP_AUTH_TOKEN validates)
+_current_mcp_authenticated: ContextVar[bool] = ContextVar("current_mcp_authenticated", default=False)
+
 
 def get_current_bank_id() -> str | None:
     """Get the current bank_id from context."""
@@ -104,6 +72,11 @@ def get_current_api_key_id() -> str | None:
     return _current_api_key_id.get()
 
 
+def get_current_mcp_authenticated() -> bool:
+    """Get whether the request was pre-authenticated by MCP transport auth."""
+    return _current_mcp_authenticated.get()
+
+
 def create_mcp_server(memory: MemoryEngine, multi_bank: bool = True) -> FastMCP:
     """
     Create and configure the Hindsight MCP server.
@@ -124,6 +97,7 @@ def create_mcp_server(memory: MemoryEngine, multi_bank: bool = True) -> FastMCP:
     _SINGLE_BANK_TOOLS: frozenset[str] = frozenset(
         {
             "retain",
+            "sync_retain",
             "recall",
             "reflect",
             "list_mental_models",
@@ -164,6 +138,7 @@ def create_mcp_server(memory: MemoryEngine, multi_bank: bool = True) -> FastMCP:
         api_key_resolver=get_current_api_key,  # Propagate API key for tenant auth
         tenant_id_resolver=get_current_tenant_id,  # Propagate tenant_id for usage metering
         api_key_id_resolver=get_current_api_key_id,  # Propagate api_key_id for usage metering
+        mcp_authenticated_resolver=get_current_mcp_authenticated,  # Propagate MCP pre-auth flag
         include_bank_id_param=multi_bank,
         tools=base_tools,
     )
@@ -182,24 +157,65 @@ def create_mcp_server(memory: MemoryEngine, multi_bank: bool = True) -> FastMCP:
     return mcp
 
 
+def _get_mcp_tools(mcp: FastMCP) -> dict:
+    """Get tool name→object mapping, compatible with FastMCP 2.x and 3.x."""
+    # FastMCP 2.x: _tool_manager._tools
+    if hasattr(mcp, "_tool_manager"):
+        return mcp._tool_manager._tools  # type: ignore[union-attr]
+    # FastMCP 3.x: _local_provider._components with "tool:" prefix
+    if hasattr(mcp, "_local_provider"):
+        return {
+            k.split(":")[1].split("@")[0]: v
+            for k, v in mcp._local_provider._components.items()  # type: ignore[union-attr]
+            if k.startswith("tool:")
+        }
+    msg = "Cannot locate tools on FastMCP instance"
+    raise AttributeError(msg)
+
+
 def _make_tools_tolerant(mcp: FastMCP) -> None:
-    """Wrap all tool run methods to strip unknown arguments before validation.
+    """Wrap all tool run methods to strip unknown arguments and coerce string-encoded JSON.
 
     LLMs frequently add extra fields like "explanation" or "reasoning" to tool calls.
     FastMCP's Pydantic TypeAdapter rejects these with "Unexpected keyword argument".
-    This wraps each tool's run() to filter arguments to only known parameters.
+
+    LLMs also frequently serialize list/dict arguments as JSON strings instead of native
+    types (e.g., tags='["a","b"]' instead of tags=["a","b"]). This auto-coerces them.
+
+    This wraps each tool's run() to apply both fixes before validation.
     """
     try:
-        for name, tool in mcp._tool_manager._tools.items():
+        tools = _get_mcp_tools(mcp)
+        for name, tool in tools.items():
             if hasattr(tool, "parameters") and tool.parameters:
-                allowed = set(tool.parameters.get("properties", {}).keys())
+                properties = tool.parameters.get("properties", {})
+                allowed = set(properties.keys())
+
+                # Build sets of parameter names that expect array or object types.
+                # Handles both direct types {"type": "array"} and anyOf/oneOf unions
+                # like {"anyOf": [{"type": "array", ...}, {"type": "null"}]}.
+                array_params: set[str] = set()
+                object_params: set[str] = set()
+                for param_name, param_schema in properties.items():
+                    _collect_coercible_types(param_schema, param_name, array_params, object_params)
+
                 original_run = tool.run
 
-                async def _tolerant_run(arguments, _allowed=allowed, _orig=original_run):
+                async def _tolerant_run(
+                    arguments,
+                    _allowed=allowed,
+                    _orig=original_run,
+                    _array_params=array_params,
+                    _object_params=object_params,
+                ):
                     extra_keys = set(arguments.keys()) - _allowed
                     if extra_keys:
                         logger.debug(f"Stripping unknown arguments from tool call: {extra_keys}")
                         arguments = {k: v for k, v in arguments.items() if k in _allowed}
+
+                    # Coerce string-encoded JSON for list/dict parameters
+                    arguments = _coerce_string_json(arguments, _array_params, _object_params)
+
                     return await _orig(arguments)
 
                 # FunctionTool is a Pydantic model with extra='forbid', so use
@@ -209,6 +225,59 @@ def _make_tools_tolerant(mcp: FastMCP) -> None:
         logger.warning(f"Could not make tools tolerant of extra arguments: {e}")
 
 
+def _collect_coercible_types(schema: dict, param_name: str, array_params: set[str], object_params: set[str]) -> None:
+    """Check a JSON Schema property and add param_name to array_params/object_params if applicable."""
+    # Direct type
+    schema_type = schema.get("type")
+    if schema_type == "array":
+        array_params.add(param_name)
+        return
+    if schema_type == "object":
+        object_params.add(param_name)
+        return
+
+    # anyOf / oneOf unions (e.g., list[str] | None → {"anyOf": [{"type": "array"}, {"type": "null"}]})
+    for variant in schema.get("anyOf", []) + schema.get("oneOf", []):
+        variant_type = variant.get("type")
+        if variant_type == "array":
+            array_params.add(param_name)
+            return
+        if variant_type == "object":
+            object_params.add(param_name)
+            return
+
+
+def _coerce_string_json(arguments: dict, array_params: set[str], object_params: set[str]) -> dict:
+    """Auto-coerce string-encoded JSON arrays/objects to native types.
+
+    LLM agents frequently serialize list and dict tool arguments as JSON strings.
+    This is backward-compatible: native arrays/objects pass through unchanged.
+    """
+    for param_name in array_params:
+        val = arguments.get(param_name)
+        if isinstance(val, str):
+            try:
+                parsed = json.loads(val)
+                if isinstance(parsed, list):
+                    arguments = {**arguments, param_name: parsed}
+                    logger.debug(f"Coerced string to list for parameter '{param_name}'")
+            except (json.JSONDecodeError, TypeError):
+                pass
+
+    for param_name in object_params:
+        val = arguments.get(param_name)
+        if isinstance(val, str):
+            try:
+                parsed = json.loads(val)
+                if isinstance(parsed, dict):
+                    arguments = {**arguments, param_name: parsed}
+                    logger.debug(f"Coerced string to dict for parameter '{param_name}'")
+            except (json.JSONDecodeError, TypeError):
+                pass
+
+    return arguments
+
+
 class MCPMiddleware:
     """ASGI middleware that intercepts MCP requests and routes to appropriate MCP server.
 
@@ -272,10 +341,12 @@ class MCPMiddleware:
             self.single_bank_server = single_bank_server
         else:
             # Create servers internally (for direct construction / tests)
+            global_config = _get_raw_config()
+            stateless = global_config.mcp_stateless
             self.multi_bank_server = create_mcp_server(memory, multi_bank=True)
-            self.multi_bank_app = self.multi_bank_server.http_app(path="/", stateless_http=True)
+            self.multi_bank_app = self.multi_bank_server.http_app(path="/", stateless_http=stateless)
             self.single_bank_server = create_mcp_server(memory, multi_bank=False)
-            self.single_bank_app = self.single_bank_server.http_app(path="/", stateless_http=True)
+            self.single_bank_app = self.single_bank_server.http_app(path="/", stateless_http=stateless)
 
     def _get_header(self, scope: dict, name: str) -> str | None:
         """Extract a header value from ASGI scope."""
@@ -298,6 +369,17 @@ class MCPMiddleware:
             await self.app(scope, receive, send)
             return
 
+        # Handle GET-before-POST gracefully (Claude Code v2.1.84+ sends GET probe before POST initialize).
+        # Without a valid Mcp-Session-Id, GET has no meaningful response — return 200 OK so
+        # the client proceeds to POST initialize instead of marking the server as failed.
+        method = scope.get("method", "")
+        if method == "GET":
+            session_id = self._get_header(scope, "Mcp-Session-Id")
+            if not session_id:
+                logger.debug("MCP GET without session ID (client probe) — returning 200 OK")
+                await self._send_ok(send)
+                return
+
         # Strip prefix from path
         path = path[len(self.prefix) :] or "/"
 
@@ -312,6 +394,7 @@ class MCPMiddleware:
         tenant_context = None
         auth_tenant_id: str | None = None
         auth_api_key_id: str | None = None
+        mcp_pre_authenticated = False
         if MCP_AUTH_TOKEN:
             # Legacy authentication mode - validate against static token
             if not auth_token:
@@ -320,8 +403,9 @@ class MCPMiddleware:
             if auth_token != MCP_AUTH_TOKEN:
                 await self._send_error(send, 401, "Invalid authentication token")
                 return
-            # Legacy mode doesn't use tenant schemas
+            # Legacy mode: mark as pre-authenticated so tenant extension won't re-validate
             tenant_context = None
+            mcp_pre_authenticated = True
         else:
             # Use TenantExtension.authenticate_mcp() for auth
             try:
@@ -368,19 +452,30 @@ class MCPMiddleware:
         # - Header/env bank_id → multi-bank app (bank_id param, all tools)
         target_app = self.single_bank_app if bank_id_from_path else self.multi_bank_app
 
-        # Set bank_id, api_key, tenant_id, and api_key_id context
+        # Set bank_id, api_key, tenant_id, api_key_id, and mcp_authenticated context
         bank_id_token = _current_bank_id.set(bank_id)
         # Store the auth token for tenant extension to validate
         api_key_token = _current_api_key.set(auth_token) if auth_token else None
         # Store tenant_id and api_key_id from authentication for usage metering
         tenant_id_token = _current_tenant_id.set(auth_tenant_id) if auth_tenant_id else None
         api_key_id_token = _current_api_key_id.set(auth_api_key_id) if auth_api_key_id else None
+        # Store MCP pre-authentication flag to skip tenant re-validation
+        mcp_auth_token = _current_mcp_authenticated.set(mcp_pre_authenticated)
         try:
             new_scope = scope.copy()
             new_scope["path"] = new_path
             # Clear root_path since we're passing directly to the app
             new_scope["root_path"] = ""
 
+            # Ensure Accept header includes required MIME types for MCP SDK.
+            # Some clients (e.g., Claude Code) don't send Accept, causing
+            # the SDK to reject with 406 Not Acceptable.
+            accept_header = self._get_header(new_scope, "accept")
+            if not accept_header or "text/event-stream" not in accept_header:
+                headers = [(k, v) for k, v in new_scope.get("headers", []) if k.lower() != b"accept"]
+                headers.append((b"accept", b"application/json, text/event-stream"))
+                new_scope["headers"] = headers
+
             # Wrap send to rewrite the SSE endpoint URL to include bank_id if using path-based routing.
             # Only rewrite SSE (text/event-stream) responses to avoid corrupting tool results
             # that might contain the literal string "data: /messages".
@@ -410,9 +505,26 @@ class MCPMiddleware:
                 _current_tenant_id.reset(tenant_id_token)
             if api_key_id_token is not None:
                 _current_api_key_id.reset(api_key_id_token)
+            _current_mcp_authenticated.reset(mcp_auth_token)
             if schema_token is not None:
                 _current_schema.reset(schema_token)
 
+    async def _send_ok(self, send):
+        """Send a 200 OK response with empty body (used for GET probes without session)."""
+        await send(
+            {
+                "type": "http.response.start",
+                "status": 200,
+                "headers": [(b"content-type", b"application/json")],
+            }
+        )
+        await send(
+            {
+                "type": "http.response.body",
+                "body": b"{}",
+            }
+        )
+
     async def _send_error(self, send, status: int, message: str, extra_headers: dict[str, str] | None = None):
         """Send an error response."""
         body = json.dumps({"error": message}).encode()
@@ -443,10 +555,14 @@ def create_mcp_servers(memory: MemoryEngine):
     Returns:
         Tuple of (multi_bank_server, single_bank_server, multi_bank_app, single_bank_app)
     """
+    global_config = _get_raw_config()
+    stateless = global_config.mcp_stateless
+
     multi_bank_server = create_mcp_server(memory, multi_bank=True)
-    multi_bank_app = multi_bank_server.http_app(path="/", stateless_http=True)
+    multi_bank_app = multi_bank_server.http_app(path="/", stateless_http=stateless)
 
     single_bank_server = create_mcp_server(memory, multi_bank=False)
-    single_bank_app = single_bank_server.http_app(path="/", stateless_http=True)
+    single_bank_app = single_bank_server.http_app(path="/", stateless_http=stateless)
 
+    logger.info(f"MCP servers created (stateless_http={stateless})")
     return multi_bank_server, single_bank_server, multi_bank_app, single_bank_app

hindsight-api-slim/hindsight_api/config.py

@@ -118,6 +118,7 @@ def normalize_config_dict(config: dict[str, Any]) -> dict[str, Any]:
 
 # Environment variable names
 ENV_DATABASE_URL = "HINDSIGHT_API_DATABASE_URL"
+ENV_MIGRATION_DATABASE_URL = "HINDSIGHT_API_MIGRATION_DATABASE_URL"
 ENV_DATABASE_SCHEMA = "HINDSIGHT_API_DATABASE_SCHEMA"
 ENV_LLM_PROVIDER = "HINDSIGHT_API_LLM_PROVIDER"
 ENV_LLM_API_KEY = "HINDSIGHT_API_LLM_API_KEY"
@@ -130,10 +131,12 @@ ENV_LLM_MAX_BACKOFF = "HINDSIGHT_API_LLM_MAX_BACKOFF"
 ENV_LLM_TIMEOUT = "HINDSIGHT_API_LLM_TIMEOUT"
 ENV_LLM_GROQ_SERVICE_TIER = "HINDSIGHT_API_LLM_GROQ_SERVICE_TIER"
 ENV_LLM_OPENAI_SERVICE_TIER = "HINDSIGHT_API_LLM_OPENAI_SERVICE_TIER"
+ENV_LLM_EXTRA_BODY = "HINDSIGHT_API_LLM_EXTRA_BODY"
 
 # Defaults for service tiers
 DEFAULT_LLM_GROQ_SERVICE_TIER = "auto"  # "on_demand", "flex", or "auto"
 DEFAULT_LLM_OPENAI_SERVICE_TIER = None  # None (default) or "flex" (50% cheaper)
+DEFAULT_LLM_EXTRA_BODY = None  # None = no extra body params; JSON dict merged into OpenAI extra_body
 
 # Per-operation LLM configuration (optional, falls back to global LLM config)
 ENV_RETAIN_LLM_PROVIDER = "HINDSIGHT_API_RETAIN_LLM_PROVIDER"
@@ -175,6 +178,14 @@ ENV_EMBEDDINGS_OPENAI_API_KEY = "HINDSIGHT_API_EMBEDDINGS_OPENAI_API_KEY"
 ENV_EMBEDDINGS_OPENAI_MODEL = "HINDSIGHT_API_EMBEDDINGS_OPENAI_MODEL"
 ENV_EMBEDDINGS_OPENAI_BASE_URL = "HINDSIGHT_API_EMBEDDINGS_OPENAI_BASE_URL"
 
+# Gemini/Vertex AI embeddings configuration
+ENV_EMBEDDINGS_GEMINI_API_KEY = "HINDSIGHT_API_EMBEDDINGS_GEMINI_API_KEY"
+ENV_EMBEDDINGS_GEMINI_MODEL = "HINDSIGHT_API_EMBEDDINGS_GEMINI_MODEL"
+ENV_EMBEDDINGS_GEMINI_OUTPUT_DIMENSIONALITY = "HINDSIGHT_API_EMBEDDINGS_GEMINI_OUTPUT_DIMENSIONALITY"
+ENV_EMBEDDINGS_VERTEXAI_PROJECT_ID = "HINDSIGHT_API_EMBEDDINGS_VERTEXAI_PROJECT_ID"
+ENV_EMBEDDINGS_VERTEXAI_REGION = "HINDSIGHT_API_EMBEDDINGS_VERTEXAI_REGION"
+ENV_EMBEDDINGS_VERTEXAI_SERVICE_ACCOUNT_KEY = "HINDSIGHT_API_EMBEDDINGS_VERTEXAI_SERVICE_ACCOUNT_KEY"
+
 # Cohere configuration (separate for embeddings and reranker)
 ENV_EMBEDDINGS_COHERE_API_KEY = "HINDSIGHT_API_EMBEDDINGS_COHERE_API_KEY"
 ENV_EMBEDDINGS_COHERE_MODEL = "HINDSIGHT_API_EMBEDDINGS_COHERE_MODEL"
@@ -183,6 +194,13 @@ ENV_RERANKER_COHERE_API_KEY = "HINDSIGHT_API_RERANKER_COHERE_API_KEY"
 ENV_RERANKER_COHERE_MODEL = "HINDSIGHT_API_RERANKER_COHERE_MODEL"
 ENV_RERANKER_COHERE_BASE_URL = "HINDSIGHT_API_RERANKER_COHERE_BASE_URL"
 
+# OpenRouter configuration (embeddings and reranker)
+ENV_OPENROUTER_API_KEY = "HINDSIGHT_API_OPENROUTER_API_KEY"
+ENV_EMBEDDINGS_OPENROUTER_API_KEY = "HINDSIGHT_API_EMBEDDINGS_OPENROUTER_API_KEY"
+ENV_EMBEDDINGS_OPENROUTER_MODEL = "HINDSIGHT_API_EMBEDDINGS_OPENROUTER_MODEL"
+ENV_RERANKER_OPENROUTER_API_KEY = "HINDSIGHT_API_RERANKER_OPENROUTER_API_KEY"
+ENV_RERANKER_OPENROUTER_MODEL = "HINDSIGHT_API_RERANKER_OPENROUTER_MODEL"
+
 # Deprecated: Legacy shared Cohere API key (for backward compatibility)
 ENV_COHERE_API_KEY = "HINDSIGHT_API_COHERE_API_KEY"
 
@@ -199,6 +217,8 @@ ENV_RERANKER_LITELLM_MAX_TOKENS_PER_DOC = "HINDSIGHT_API_RERANKER_LITELLM_MAX_TO
 ENV_EMBEDDINGS_LITELLM_SDK_API_KEY = "HINDSIGHT_API_EMBEDDINGS_LITELLM_SDK_API_KEY"
 ENV_EMBEDDINGS_LITELLM_SDK_MODEL = "HINDSIGHT_API_EMBEDDINGS_LITELLM_SDK_MODEL"
 ENV_EMBEDDINGS_LITELLM_SDK_API_BASE = "HINDSIGHT_API_EMBEDDINGS_LITELLM_SDK_API_BASE"
+ENV_EMBEDDINGS_LITELLM_SDK_OUTPUT_DIMENSIONS = "HINDSIGHT_API_EMBEDDINGS_LITELLM_SDK_OUTPUT_DIMENSIONS"
+ENV_EMBEDDINGS_LITELLM_SDK_ENCODING_FORMAT = "HINDSIGHT_API_EMBEDDINGS_LITELLM_SDK_ENCODING_FORMAT"
 ENV_RERANKER_LITELLM_SDK_API_KEY = "HINDSIGHT_API_RERANKER_LITELLM_SDK_API_KEY"
 ENV_RERANKER_LITELLM_SDK_MODEL = "HINDSIGHT_API_RERANKER_LITELLM_SDK_MODEL"
 ENV_RERANKER_LITELLM_SDK_API_BASE = "HINDSIGHT_API_RERANKER_LITELLM_SDK_API_BASE"
@@ -212,6 +232,9 @@ ENV_RERANKER_LOCAL_MODEL = "HINDSIGHT_API_RERANKER_LOCAL_MODEL"
 ENV_RERANKER_LOCAL_FORCE_CPU = "HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU"
 ENV_RERANKER_LOCAL_MAX_CONCURRENT = "HINDSIGHT_API_RERANKER_LOCAL_MAX_CONCURRENT"
 ENV_RERANKER_LOCAL_TRUST_REMOTE_CODE = "HINDSIGHT_API_RERANKER_LOCAL_TRUST_REMOTE_CODE"
+ENV_RERANKER_LOCAL_FP16 = "HINDSIGHT_API_RERANKER_LOCAL_FP16"
+ENV_RERANKER_LOCAL_BUCKET_BATCHING = "HINDSIGHT_API_RERANKER_LOCAL_BUCKET_BATCHING"
+ENV_RERANKER_LOCAL_BATCH_SIZE = "HINDSIGHT_API_RERANKER_LOCAL_BATCH_SIZE"
 ENV_RERANKER_TEI_URL = "HINDSIGHT_API_RERANKER_TEI_URL"
 ENV_RERANKER_TEI_BATCH_SIZE = "HINDSIGHT_API_RERANKER_TEI_BATCH_SIZE"
 ENV_RERANKER_TEI_MAX_CONCURRENT = "HINDSIGHT_API_RERANKER_TEI_MAX_CONCURRENT"
@@ -222,6 +245,17 @@ ENV_RERANKER_FLASHRANK_CACHE_DIR = "HINDSIGHT_API_RERANKER_FLASHRANK_CACHE_DIR"
 # ZeroEntropy configuration (reranker only)
 ENV_RERANKER_ZEROENTROPY_API_KEY = "HINDSIGHT_API_RERANKER_ZEROENTROPY_API_KEY"
 ENV_RERANKER_ZEROENTROPY_MODEL = "HINDSIGHT_API_RERANKER_ZEROENTROPY_MODEL"
+ENV_RERANKER_ZEROENTROPY_BASE_URL = "HINDSIGHT_API_RERANKER_ZEROENTROPY_BASE_URL"
+
+# SiliconFlow configuration (reranker only; Cohere-compatible /rerank endpoint)
+ENV_RERANKER_SILICONFLOW_API_KEY = "HINDSIGHT_API_RERANKER_SILICONFLOW_API_KEY"
+ENV_RERANKER_SILICONFLOW_MODEL = "HINDSIGHT_API_RERANKER_SILICONFLOW_MODEL"
+ENV_RERANKER_SILICONFLOW_BASE_URL = "HINDSIGHT_API_RERANKER_SILICONFLOW_BASE_URL"
+
+# Google Discovery Engine reranker configuration
+ENV_RERANKER_GOOGLE_MODEL = "HINDSIGHT_API_RERANKER_GOOGLE_MODEL"
+ENV_RERANKER_GOOGLE_PROJECT_ID = "HINDSIGHT_API_RERANKER_GOOGLE_PROJECT_ID"
+ENV_RERANKER_GOOGLE_SERVICE_ACCOUNT_KEY = "HINDSIGHT_API_RERANKER_GOOGLE_SERVICE_ACCOUNT_KEY"
 
 ENV_VECTOR_EXTENSION = "HINDSIGHT_API_VECTOR_EXTENSION"
 ENV_TEXT_SEARCH_EXTENSION = "HINDSIGHT_API_TEXT_SEARCH_EXTENSION"
@@ -234,13 +268,16 @@ ENV_LOG_FORMAT = "HINDSIGHT_API_LOG_FORMAT"
 ENV_WORKERS = "HINDSIGHT_API_WORKERS"
 ENV_MCP_ENABLED = "HINDSIGHT_API_MCP_ENABLED"
 ENV_MCP_ENABLED_TOOLS = "HINDSIGHT_API_MCP_ENABLED_TOOLS"
+ENV_MCP_STATELESS = "HINDSIGHT_API_MCP_STATELESS"
 ENV_ENABLE_BANK_CONFIG_API = "HINDSIGHT_API_ENABLE_BANK_CONFIG_API"
+ENV_DEFAULT_BANK_TEMPLATE = "HINDSIGHT_API_DEFAULT_BANK_TEMPLATE"
 ENV_GRAPH_RETRIEVER = "HINDSIGHT_API_GRAPH_RETRIEVER"
-ENV_MPFP_TOP_K_NEIGHBORS = "HINDSIGHT_API_MPFP_TOP_K_NEIGHBORS"
 ENV_RECALL_MAX_CONCURRENT = "HINDSIGHT_API_RECALL_MAX_CONCURRENT"
 ENV_RECALL_CONNECTION_BUDGET = "HINDSIGHT_API_RECALL_CONNECTION_BUDGET"
 ENV_RECALL_MAX_QUERY_TOKENS = "HINDSIGHT_API_RECALL_MAX_QUERY_TOKENS"
 ENV_MENTAL_MODEL_REFRESH_CONCURRENCY = "HINDSIGHT_API_MENTAL_MODEL_REFRESH_CONCURRENCY"
+ENV_LINK_EXPANSION_PER_ENTITY_LIMIT = "HINDSIGHT_API_LINK_EXPANSION_PER_ENTITY_LIMIT"
+ENV_LINK_EXPANSION_TIMEOUT = "HINDSIGHT_API_LINK_EXPANSION_TIMEOUT"
 
 # OpenTelemetry tracing configuration
 ENV_OTEL_TRACES_ENABLED = "HINDSIGHT_API_OTEL_TRACES_ENABLED"
@@ -248,6 +285,7 @@ ENV_OTEL_EXPORTER_OTLP_ENDPOINT = "HINDSIGHT_API_OTEL_EXPORTER_OTLP_ENDPOINT"
 ENV_OTEL_EXPORTER_OTLP_HEADERS = "HINDSIGHT_API_OTEL_EXPORTER_OTLP_HEADERS"
 ENV_OTEL_SERVICE_NAME = "HINDSIGHT_API_OTEL_SERVICE_NAME"
 ENV_OTEL_DEPLOYMENT_ENVIRONMENT = "HINDSIGHT_API_OTEL_DEPLOYMENT_ENVIRONMENT"
+ENV_METRICS_INCLUDE_BANK_ID = "HINDSIGHT_API_METRICS_INCLUDE_BANK_ID"
 
 # Vertex AI configuration
 ENV_LLM_VERTEXAI_PROJECT_ID = "HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID"
@@ -264,10 +302,12 @@ ENV_RETAIN_EXTRACT_CAUSAL_LINKS = "HINDSIGHT_API_RETAIN_EXTRACT_CAUSAL_LINKS"
 ENV_RETAIN_EXTRACTION_MODE = "HINDSIGHT_API_RETAIN_EXTRACTION_MODE"
 ENV_RETAIN_MISSION = "HINDSIGHT_API_RETAIN_MISSION"
 ENV_RETAIN_CUSTOM_INSTRUCTIONS = "HINDSIGHT_API_RETAIN_CUSTOM_INSTRUCTIONS"
+ENV_RETAIN_DEFAULT_STRATEGY = "HINDSIGHT_API_RETAIN_DEFAULT_STRATEGY"
 ENV_RETAIN_BATCH_TOKENS = "HINDSIGHT_API_RETAIN_BATCH_TOKENS"
 ENV_RETAIN_ENTITY_LOOKUP = "HINDSIGHT_API_RETAIN_ENTITY_LOOKUP"
 ENV_RETAIN_BATCH_ENABLED = "HINDSIGHT_API_RETAIN_BATCH_ENABLED"
 ENV_RETAIN_BATCH_POLL_INTERVAL_SECONDS = "HINDSIGHT_API_RETAIN_BATCH_POLL_INTERVAL_SECONDS"
+ENV_RETAIN_CHUNK_BATCH_SIZE = "HINDSIGHT_API_RETAIN_CHUNK_BATCH_SIZE"
 
 # File storage configuration
 ENV_FILE_STORAGE_TYPE = "HINDSIGHT_API_FILE_STORAGE_TYPE"
@@ -300,6 +340,7 @@ ENV_CONSOLIDATION_SOURCE_FACTS_MAX_TOKENS_PER_OBSERVATION = (
     "HINDSIGHT_API_CONSOLIDATION_SOURCE_FACTS_MAX_TOKENS_PER_OBSERVATION"
 )
 ENV_OBSERVATIONS_MISSION = "HINDSIGHT_API_OBSERVATIONS_MISSION"
+ENV_MAX_OBSERVATIONS_PER_SCOPE = "HINDSIGHT_API_MAX_OBSERVATIONS_PER_SCOPE"
 ENV_ENABLE_OBSERVATION_HISTORY = "HINDSIGHT_API_ENABLE_OBSERVATION_HISTORY"
 ENV_ENABLE_MENTAL_MODEL_HISTORY = "HINDSIGHT_API_ENABLE_MENTAL_MODEL_HISTORY"
 
@@ -309,6 +350,14 @@ ENV_WEBHOOK_SECRET = "HINDSIGHT_API_WEBHOOK_SECRET"
 ENV_WEBHOOK_EVENT_TYPES = "HINDSIGHT_API_WEBHOOK_EVENT_TYPES"
 ENV_WEBHOOK_DELIVERY_POLL_INTERVAL_SECONDS = "HINDSIGHT_API_WEBHOOK_DELIVERY_POLL_INTERVAL_SECONDS"
 
+# Built-in llama.cpp configuration (for provider=llamacpp)
+ENV_LLAMACPP_MODEL_PATH = "HINDSIGHT_API_LLAMACPP_MODEL_PATH"
+ENV_LLAMACPP_GPU_LAYERS = "HINDSIGHT_API_LLAMACPP_GPU_LAYERS"
+ENV_LLAMACPP_CONTEXT_SIZE = "HINDSIGHT_API_LLAMACPP_CONTEXT_SIZE"
+ENV_LLAMACPP_CHAT_FORMAT = "HINDSIGHT_API_LLAMACPP_CHAT_FORMAT"
+ENV_LLAMACPP_NO_GRAMMAR = "HINDSIGHT_API_LLAMACPP_NO_GRAMMAR"
+ENV_LLAMACPP_EXTRA_ARGS = "HINDSIGHT_API_LLAMACPP_EXTRA_ARGS"
+
 # Optimization flags
 ENV_SKIP_LLM_VERIFICATION = "HINDSIGHT_API_SKIP_LLM_VERIFICATION"
 ENV_LAZY_RERANKER = "HINDSIGHT_API_LAZY_RERANKER"
@@ -330,11 +379,22 @@ ENV_WORKER_MAX_RETRIES = "HINDSIGHT_API_WORKER_MAX_RETRIES"
 ENV_WORKER_HTTP_PORT = "HINDSIGHT_API_WORKER_HTTP_PORT"
 ENV_WORKER_MAX_SLOTS = "HINDSIGHT_API_WORKER_MAX_SLOTS"
 ENV_WORKER_CONSOLIDATION_MAX_SLOTS = "HINDSIGHT_API_WORKER_CONSOLIDATION_MAX_SLOTS"
+ENV_RETAIN_MAX_CONCURRENT = "HINDSIGHT_API_RETAIN_MAX_CONCURRENT"
 
 # Reflect agent settings
 ENV_REFLECT_MAX_ITERATIONS = "HINDSIGHT_API_REFLECT_MAX_ITERATIONS"
 ENV_REFLECT_MAX_CONTEXT_TOKENS = "HINDSIGHT_API_REFLECT_MAX_CONTEXT_TOKENS"
+ENV_REFLECT_WALL_TIMEOUT = "HINDSIGHT_API_REFLECT_WALL_TIMEOUT"
 ENV_REFLECT_MISSION = "HINDSIGHT_API_REFLECT_MISSION"
+ENV_REFLECT_SOURCE_FACTS_MAX_TOKENS = "HINDSIGHT_API_REFLECT_SOURCE_FACTS_MAX_TOKENS"
+ENV_RECALL_INCLUDE_CHUNKS = "HINDSIGHT_API_RECALL_INCLUDE_CHUNKS"
+ENV_RECALL_MAX_TOKENS = "HINDSIGHT_API_RECALL_MAX_TOKENS"
+ENV_RECALL_CHUNKS_MAX_TOKENS = "HINDSIGHT_API_RECALL_CHUNKS_MAX_TOKENS"
+
+# Audit log settings
+ENV_AUDIT_LOG_ENABLED = "HINDSIGHT_API_AUDIT_LOG_ENABLED"
+ENV_AUDIT_LOG_ACTIONS = "HINDSIGHT_API_AUDIT_LOG_ACTIONS"
+ENV_AUDIT_LOG_RETENTION_DAYS = "HINDSIGHT_API_AUDIT_LOG_RETENTION_DAYS"
 
 # Disposition settings
 ENV_DISPOSITION_SKEPTICISM = "HINDSIGHT_API_DISPOSITION_SKEPTICISM"
@@ -349,18 +409,31 @@ DEFAULT_LLM_PROVIDER = "openai"
 # Provider-specific default models
 PROVIDER_DEFAULT_MODELS = {
     "openai": "gpt-4o-mini",
-    "anthropic": "claude-haiku-4-5-20251001",
+    "anthropic": "claude-haiku-4-5",
     "gemini": "gemini-2.5-flash",
     "groq": "openai/gpt-oss-120b",
-    "minimax": "MiniMax-M2.5",
+    "minimax": "MiniMax-M2.7",
     "ollama": "gemma3:12b",
+    "llamacpp": "gemma-4-e2b-it",
     "lmstudio": "local-model",
     "vertexai": "google/gemini-2.5-flash-lite",
     "openai-codex": "gpt-5.2-codex",
     "claude-code": "claude-sonnet-4-5-20250929",
     "mock": "mock-model",
+    "none": "none",
+    "litellm": "gpt-4o-mini",
+    "bedrock": "us.amazon.nova-2-lite-v1:0",
+    "volcano": "doubao-pro-32k",
+    "openrouter": "qwen/qwen3.5-9b",
 }
 DEFAULT_LLM_MODEL = "gpt-4o-mini"  # Fallback if provider not in table
+# Built-in llama.cpp defaults
+DEFAULT_LLAMACPP_GPU_LAYERS = -1  # -1 = offload all layers to GPU (Metal/CUDA)
+DEFAULT_LLAMACPP_CONTEXT_SIZE = 8192
+DEFAULT_LLAMACPP_CHAT_FORMAT = None  # None = auto-detect from GGUF metadata
+DEFAULT_LLAMACPP_NO_GRAMMAR = False  # True = disable JSON grammar enforcement (faster but less reliable)
+DEFAULT_LLAMACPP_EXTRA_ARGS = None  # Space-separated extra CLI args for llama.cpp server
+
 DEFAULT_LLM_MAX_CONCURRENT = 32
 DEFAULT_LLM_MAX_RETRIES = 10  # Max retry attempts for LLM API calls
 DEFAULT_LLM_INITIAL_BACKOFF = 1.0  # Initial backoff in seconds for retry exponential backoff
@@ -380,6 +453,8 @@ DEFAULT_EMBEDDINGS_LOCAL_MODEL = "BAAI/bge-small-en-v1.5"
 DEFAULT_EMBEDDINGS_LOCAL_FORCE_CPU = False  # Force CPU mode for local embeddings (avoids MPS/XPC issues on macOS)
 DEFAULT_EMBEDDINGS_LOCAL_TRUST_REMOTE_CODE = False  # Security: disabled by default, required for some models
 DEFAULT_EMBEDDINGS_OPENAI_MODEL = "text-embedding-3-small"
+DEFAULT_EMBEDDINGS_GEMINI_MODEL = "gemini-embedding-001"
+DEFAULT_EMBEDDINGS_GEMINI_OUTPUT_DIMENSIONALITY = 768
 DEFAULT_EMBEDDING_DIMENSION = 384
 
 DEFAULT_RERANKER_PROVIDER = "local"
@@ -389,6 +464,9 @@ DEFAULT_RERANKER_LOCAL_MAX_CONCURRENT = 4  # Limit concurrent CPU-bound rerankin
 DEFAULT_RERANKER_LOCAL_TRUST_REMOTE_CODE = (
     False  # Security: disabled by default, required for some models like jina-reranker-v2
 )
+DEFAULT_RERANKER_LOCAL_FP16 = False  # FP16 inference: opt-in, faster on MPS/CUDA (not CPU)
+DEFAULT_RERANKER_LOCAL_BUCKET_BATCHING = False  # Length-sorted bucket batching: opt-in, 36-54% speedup
+DEFAULT_RERANKER_LOCAL_BATCH_SIZE = 32  # Batch size for local reranker predict() calls
 DEFAULT_RERANKER_TEI_BATCH_SIZE = 128
 DEFAULT_RERANKER_TEI_MAX_CONCURRENT = 8
 DEFAULT_RERANKER_MAX_CANDIDATES = 300
@@ -398,8 +476,17 @@ DEFAULT_RERANKER_FLASHRANK_CACHE_DIR = None  # Use default cache directory
 DEFAULT_EMBEDDINGS_COHERE_MODEL = "embed-english-v3.0"
 DEFAULT_RERANKER_COHERE_MODEL = "rerank-english-v3.0"
 
+# OpenRouter defaults
+DEFAULT_EMBEDDINGS_OPENROUTER_MODEL = "perplexity/pplx-embed-v1-0.6b"
+DEFAULT_RERANKER_OPENROUTER_MODEL = "cohere/rerank-v3.5"
+
 DEFAULT_RERANKER_ZEROENTROPY_MODEL = "zerank-2"
 
+DEFAULT_RERANKER_SILICONFLOW_MODEL = "BAAI/bge-reranker-v2-m3"
+DEFAULT_RERANKER_SILICONFLOW_BASE_URL = "https://api.siliconflow.cn/v1"
+
+DEFAULT_RERANKER_GOOGLE_MODEL = "semantic-ranker-default-004"
+
 # Vector extension (pgvector, vchord, or pgvectorscale)
 DEFAULT_VECTOR_EXTENSION = "pgvector"  # Options: "pgvector", "vchord", "pgvectorscale"
 
@@ -414,6 +501,7 @@ DEFAULT_RERANKER_LITELLM_MAX_TOKENS_PER_DOC: int | None = None
 
 # LiteLLM SDK defaults
 DEFAULT_EMBEDDINGS_LITELLM_SDK_MODEL = "cohere/embed-english-v3.0"
+DEFAULT_EMBEDDINGS_LITELLM_SDK_ENCODING_FORMAT = "float"
 DEFAULT_RERANKER_LITELLM_SDK_MODEL = "cohere/rerank-english-v3.0"
 
 DEFAULT_HOST = "0.0.0.0"
@@ -424,22 +512,30 @@ DEFAULT_LOG_FORMAT = "text"  # Options: "text", "json"
 DEFAULT_WORKERS = 1
 DEFAULT_MCP_ENABLED = True
 DEFAULT_MCP_ENABLED_TOOLS: list[str] | None = None  # None = all tools enabled
+DEFAULT_MCP_STATELESS = False  # False = stateful (supports SSE/GET); True = stateless (POST-only)
 DEFAULT_ENABLE_BANK_CONFIG_API = True
-DEFAULT_GRAPH_RETRIEVER = "link_expansion"  # Options: "link_expansion", "mpfp", "bfs"
-DEFAULT_MPFP_TOP_K_NEIGHBORS = 20  # Fan-out limit per node in MPFP graph traversal
+DEFAULT_DEFAULT_BANK_TEMPLATE: dict | None = None  # BankTemplateManifest dict applied to newly-created banks
+DEFAULT_GRAPH_RETRIEVER = "link_expansion"
 DEFAULT_RECALL_MAX_CONCURRENT = 32  # Max concurrent recall operations per worker
 DEFAULT_RECALL_CONNECTION_BUDGET = 4  # Max concurrent DB connections per recall operation
 DEFAULT_RECALL_MAX_QUERY_TOKENS = 500  # Maximum tokens allowed in recall query
 DEFAULT_MENTAL_MODEL_REFRESH_CONCURRENCY = 8  # Max concurrent mental model refreshes
+DEFAULT_LINK_EXPANSION_PER_ENTITY_LIMIT = 200  # Max target units per entity in graph expansion
+DEFAULT_LINK_EXPANSION_TIMEOUT = 10.0  # Timeout (seconds) for entity expansion query
 
 # Retain settings
 DEFAULT_RETAIN_MAX_COMPLETION_TOKENS = 64000  # Max tokens for fact extraction LLM call
 DEFAULT_RETAIN_CHUNK_SIZE = 3000  # Max chars per chunk for fact extraction
 DEFAULT_RETAIN_EXTRACT_CAUSAL_LINKS = True  # Extract causal links between facts
 DEFAULT_RETAIN_EXTRACTION_MODE = "concise"  # Extraction mode: "concise", "verbose", or "custom"
-RETAIN_EXTRACTION_MODES = ("concise", "verbose", "custom")  # Allowed extraction modes
+RETAIN_EXTRACTION_MODES = ("concise", "verbose", "custom", "verbatim", "chunks")  # Allowed extraction modes
 DEFAULT_RETAIN_MISSION = None  # Declarative spec of what to retain (injected into any extraction mode)
 DEFAULT_RETAIN_CUSTOM_INSTRUCTIONS = None  # Custom extraction guidelines (only used when mode="custom")
+DEFAULT_RETAIN_DEFAULT_STRATEGY = None  # Default strategy name (None = no strategy override)
+DEFAULT_RETAIN_STRATEGIES: dict | None = None  # Named retain strategies (dict of name → config overrides)
+DEFAULT_RETAIN_CHUNK_BATCH_SIZE = (
+    100  # Max chunks per streaming batch. Each chunk produces ~17 facts, so 100 chunks = ~1700 facts/batch.
+)
 DEFAULT_RETAIN_BATCH_TOKENS = 10_000  # ~40KB of text  # Max chars per sub-batch for async retain auto-splitting
 DEFAULT_RETAIN_ENTITY_LOOKUP = "trigram"  # "full" or "trigram"
 DEFAULT_RETAIN_BATCH_ENABLED = False  # Use LLM Batch API for fact extraction (only when async=True)
@@ -468,6 +564,7 @@ DEFAULT_CONSOLIDATION_SOURCE_FACTS_MAX_TOKENS_PER_OBSERVATION = (
     256  # Max tokens of source facts per observation in consolidation prompt (-1 = unlimited)
 )
 DEFAULT_OBSERVATIONS_MISSION = None  # Declarative spec of what observations are for this bank
+DEFAULT_MAX_OBSERVATIONS_PER_SCOPE = -1  # Max observations per tag scope (-1 = unlimited)
 
 # Database migrations
 DEFAULT_RUN_MIGRATIONS_ON_STARTUP = True
@@ -486,10 +583,16 @@ DEFAULT_WORKER_MAX_RETRIES = 3  # Max retries before marking task failed
 DEFAULT_WORKER_HTTP_PORT = 8889  # HTTP port for worker metrics/health
 DEFAULT_WORKER_MAX_SLOTS = 10  # Total concurrent tasks per worker
 DEFAULT_WORKER_CONSOLIDATION_MAX_SLOTS = 2  # Max concurrent consolidation tasks per worker
+DEFAULT_RETAIN_MAX_CONCURRENT = 4  # Max concurrent retain DB phases (HNSW reads + writes). Limits I/O contention.
 
 # Reflect agent settings
 DEFAULT_REFLECT_MAX_ITERATIONS = 10  # Max tool call iterations before forcing response
 DEFAULT_REFLECT_MAX_CONTEXT_TOKENS = 100_000  # Max accumulated context tokens before forcing final prompt
+DEFAULT_REFLECT_WALL_TIMEOUT = 300  # Wall-clock timeout in seconds for the entire reflect operation (5 minutes)
+DEFAULT_REFLECT_SOURCE_FACTS_MAX_TOKENS = -1  # Token budget for source facts in search_observations (-1 = disabled)
+DEFAULT_RECALL_INCLUDE_CHUNKS = True  # Whether internal recall (e.g. mental model refresh) returns raw chunks
+DEFAULT_RECALL_MAX_TOKENS = 2048  # Token budget for facts returned by internal recall
+DEFAULT_RECALL_CHUNKS_MAX_TOKENS = 1000  # Token budget for raw chunks returned by internal recall
 
 # Disposition defaults (None = not set, fall back to bank DB value or 3)
 DEFAULT_DISPOSITION_SKEPTICISM = None
@@ -500,6 +603,12 @@ DEFAULT_DISPOSITION_EMPATHY = None
 DEFAULT_OTEL_TRACES_ENABLED = False  # Disabled by default for backward compatibility
 DEFAULT_OTEL_SERVICE_NAME = "hindsight-api"
 DEFAULT_OTEL_DEPLOYMENT_ENVIRONMENT = "development"
+DEFAULT_METRICS_INCLUDE_BANK_ID = False  # Disabled by default to avoid high-cardinality OTel metric growth
+
+# Audit log defaults
+DEFAULT_AUDIT_LOG_ENABLED = False  # Disabled by default
+DEFAULT_AUDIT_LOG_ACTIONS = ""  # Empty = audit all eligible actions
+DEFAULT_AUDIT_LOG_RETENTION_DAYS = -1  # -1 = keep forever
 
 # Default MCP tool descriptions (can be customized via env vars)
 DEFAULT_MCP_RETAIN_DESCRIPTION = """Store important information to long-term memory.
@@ -583,12 +692,33 @@ def _get_default_model_for_provider(provider: str) -> str:
     return PROVIDER_DEFAULT_MODELS.get(provider.lower(), DEFAULT_LLM_MODEL)
 
 
+def _parse_default_bank_template(raw: str | None) -> dict | None:
+    """
+    Parse HINDSIGHT_API_DEFAULT_BANK_TEMPLATE as JSON.
+
+    The env var holds a BankTemplateManifest (JSON object) applied verbatim to
+    every newly-created bank. Full Pydantic validation is deferred to bank
+    creation time (to avoid pulling API models into config.py), but we fail
+    fast here if the value is not valid JSON or not a JSON object.
+    """
+    if raw is None or raw.strip() == "":
+        return DEFAULT_DEFAULT_BANK_TEMPLATE
+    try:
+        parsed = json.loads(raw)
+    except json.JSONDecodeError as e:
+        raise ValueError(f"Invalid {ENV_DEFAULT_BANK_TEMPLATE}: expected a JSON object, got invalid JSON: {e}") from e
+    if not isinstance(parsed, dict):
+        raise ValueError(f"Invalid {ENV_DEFAULT_BANK_TEMPLATE}: expected a JSON object, got {type(parsed).__name__}")
+    return parsed
+
+
 @dataclass
 class HindsightConfig:
     """Configuration container for Hindsight API."""
 
     # Database
     database_url: str
+    migration_database_url: str | None
     database_schema: str
     vector_extension: str  # "pgvector" or "vchord"
     text_search_extension: str  # "native" or "vchord"
@@ -605,6 +735,9 @@ class HindsightConfig:
     llm_timeout: float
     llm_groq_service_tier: str  # Groq: "on_demand", "flex", or "auto"
     llm_openai_service_tier: str | None  # OpenAI: None (default) or "flex" (50% cheaper)
+    llm_extra_body: (
+        dict | None
+    )  # Extra body params merged into OpenAI-compatible API calls (e.g. {"chat_template_kwargs": {"enable_thinking": true}})
 
     # Vertex AI configuration
     llm_vertexai_project_id: str | None
@@ -614,6 +747,14 @@ class HindsightConfig:
     # Gemini safety settings (None = use Gemini defaults; list of dicts with category/threshold)
     llm_gemini_safety_settings: list | None
 
+    # Built-in llama.cpp configuration (for provider=llamacpp)
+    llamacpp_model_path: str | None  # Path to GGUF file (None = auto-download default)
+    llamacpp_gpu_layers: int  # -1 = all layers on GPU, 0 = CPU only
+    llamacpp_context_size: int  # Context window size
+    llamacpp_chat_format: str | None  # Chat template format (None = auto-detect from GGUF)
+    llamacpp_no_grammar: bool  # Disable JSON grammar enforcement (faster, less reliable)
+    llamacpp_extra_args: str | None  # Space-separated extra CLI args for llama.cpp server
+
     # Per-operation LLM configuration (None = use default LLM config)
     retain_llm_provider: str | None
     retain_llm_api_key: str | None
@@ -655,12 +796,23 @@ class HindsightConfig:
     embeddings_cohere_api_key: str | None
     embeddings_cohere_model: str
     embeddings_cohere_base_url: str | None
+    embeddings_openrouter_api_key: str | None
+    embeddings_openrouter_model: str
     embeddings_litellm_api_base: str
     embeddings_litellm_api_key: str | None
     embeddings_litellm_model: str
     embeddings_litellm_sdk_api_key: str | None
     embeddings_litellm_sdk_model: str
     embeddings_litellm_sdk_api_base: str | None
+    embeddings_litellm_sdk_output_dimensions: int | None
+    embeddings_litellm_sdk_encoding_format: str | None
+    # Gemini/Vertex AI embeddings
+    embeddings_gemini_api_key: str | None
+    embeddings_gemini_model: str
+    embeddings_gemini_output_dimensionality: int | None
+    embeddings_vertexai_project_id: str | None
+    embeddings_vertexai_region: str | None
+    embeddings_vertexai_service_account_key: str | None
 
     # Reranker
     reranker_provider: str
@@ -668,6 +820,9 @@ class HindsightConfig:
     reranker_local_force_cpu: bool
     reranker_local_max_concurrent: int
     reranker_local_trust_remote_code: bool
+    reranker_local_fp16: bool
+    reranker_local_bucket_batching: bool
+    reranker_local_batch_size: int
     reranker_tei_url: str | None
     reranker_tei_batch_size: int
     reranker_tei_max_concurrent: int
@@ -675,6 +830,8 @@ class HindsightConfig:
     reranker_cohere_api_key: str | None
     reranker_cohere_model: str
     reranker_cohere_base_url: str | None
+    reranker_openrouter_api_key: str | None
+    reranker_openrouter_model: str
     reranker_litellm_api_base: str
     reranker_litellm_api_key: str | None
     reranker_litellm_model: str
@@ -684,6 +841,13 @@ class HindsightConfig:
     reranker_litellm_sdk_api_base: str | None
     reranker_zeroentropy_api_key: str | None
     reranker_zeroentropy_model: str
+    reranker_zeroentropy_base_url: str | None
+    reranker_siliconflow_api_key: str | None
+    reranker_siliconflow_model: str
+    reranker_siliconflow_base_url: str
+    reranker_google_model: str
+    reranker_google_project_id: str | None
+    reranker_google_service_account_key: str | None
 
     # Server
     host: str
@@ -693,15 +857,20 @@ class HindsightConfig:
     log_format: str
     mcp_enabled: bool
     mcp_enabled_tools: list[str] | None  # None = all tools; explicit list = allowlist
+    mcp_stateless: bool  # True = stateless HTTP (POST-only); False = stateful (supports GET/SSE)
     enable_bank_config_api: bool
+    # Default bank template (static, server-level only). When set, the manifest is applied
+    # to every newly-created bank, overriding the env/config defaults for any fields it sets.
+    default_bank_template: dict | None
 
     # Recall
     graph_retriever: str
-    mpfp_top_k_neighbors: int
     recall_max_concurrent: int
     recall_connection_budget: int
     recall_max_query_tokens: int
     mental_model_refresh_concurrency: int
+    link_expansion_per_entity_limit: int
+    link_expansion_timeout: float
 
     # Retain settings
     retain_max_completion_tokens: int
@@ -710,10 +879,13 @@ class HindsightConfig:
     retain_extraction_mode: str
     retain_mission: str | None
     retain_custom_instructions: str | None
+    retain_default_strategy: str | None
+    retain_strategies: dict | None
     retain_batch_tokens: int
     retain_batch_enabled: bool
     retain_batch_poll_interval_seconds: int
     retain_entity_lookup: str  # "full" or "trigram"
+    retain_chunk_batch_size: int  # Max chunks per streaming batch (0 = disabled)
 
     # File storage (static - server-level only)
     file_storage_type: str  # "native" (PostgreSQL) or "s3" (S3-compatible)
@@ -746,6 +918,7 @@ class HindsightConfig:
     consolidation_source_facts_max_tokens: int
     consolidation_source_facts_max_tokens_per_observation: int
     observations_mission: str | None
+    max_observations_per_scope: int
 
     # Entity labels (controlled vocabulary of key:value classification labels extracted at retain time)
     # List of label group dicts: [{key, description, type, optional, values: [{value, description}]}]
@@ -756,6 +929,12 @@ class HindsightConfig:
 
     # Reflect agent settings
     reflect_mission: str | None
+    reflect_source_facts_max_tokens: int
+
+    # Recall settings (used by internal recall, e.g. during mental model refresh)
+    recall_include_chunks: bool
+    recall_max_tokens: int
+    recall_chunks_max_tokens: int
 
     # Disposition settings (hierarchical - can be overridden per bank; None = fall back to DB)
     disposition_skepticism: int | None
@@ -783,10 +962,12 @@ class HindsightConfig:
     worker_http_port: int
     worker_max_slots: int
     worker_consolidation_max_slots: int
+    retain_max_concurrent: int
 
     # Reflect agent settings
     reflect_max_iterations: int
     reflect_max_context_tokens: int
+    reflect_wall_timeout: int
 
     # OpenTelemetry tracing configuration
     otel_traces_enabled: bool
@@ -794,6 +975,12 @@ class HindsightConfig:
     otel_exporter_otlp_headers: str | None
     otel_service_name: str
     otel_deployment_environment: str
+    metrics_include_bank_id: bool
+
+    # Audit log configuration (static - server-level only)
+    audit_log_enabled: bool  # Master switch for audit logging
+    audit_log_actions: list[str]  # Allowlist of action types (empty = all)
+    audit_log_retention_days: int  # -1 = keep forever, >0 = delete after N days
 
     # Webhook configuration (static - server-level only, not per-bank)
     webhook_url: str | None  # Global webhook URL (None = disabled)
@@ -818,8 +1005,14 @@ class HindsightConfig:
         "embeddings_tei_base_url",
         "reranker_tei_base_url",
         "reranker_cohere_base_url",
+        "reranker_zeroentropy_base_url",
+        "reranker_siliconflow_base_url",
         # Service Account Keys
         "llm_vertexai_service_account_key",
+        "embeddings_vertexai_service_account_key",
+        "reranker_google_service_account_key",
+        # Embeddings API keys
+        "embeddings_gemini_api_key",
         # File storage credentials
         "file_storage_s3_access_key_id",
         "file_storage_s3_secret_access_key",
@@ -840,6 +1033,9 @@ class HindsightConfig:
         "retain_extraction_mode",
         "retain_mission",
         "retain_custom_instructions",
+        "retain_default_strategy",
+        "retain_strategies",
+        "retain_chunk_batch_size",
         # Entity labels (controlled vocabulary for entity classification)
         "entity_labels",
         "entities_allow_free_form",
@@ -849,8 +1045,14 @@ class HindsightConfig:
         "consolidation_source_facts_max_tokens",
         "consolidation_source_facts_max_tokens_per_observation",
         "observations_mission",
+        "max_observations_per_scope",
         # Reflect settings
         "reflect_mission",
+        "reflect_source_facts_max_tokens",
+        # Recall settings (used by internal recall, e.g. mental model refresh)
+        "recall_include_chunks",
+        "recall_max_tokens",
+        "recall_chunks_max_tokens",
         # Disposition settings
         "disposition_skepticism",
         "disposition_literalism",
@@ -933,9 +1135,19 @@ class HindsightConfig:
                 f"Invalid text_search_extension: {self.text_search_extension}. Must be one of: {', '.join(valid_text_search)}"
             )
 
+        # When LLM provider is "none", force chunks-only mode and disable LLM-dependent features
+        if self.llm_provider == "none":
+            self.retain_extraction_mode = "chunks"
+            self.enable_observations = False
+            logger.info(
+                "LLM provider set to 'none': forcing retain_extraction_mode='chunks', "
+                "disabling observations/consolidation. Reflect will return HTTP 400."
+            )
+
         # RETAIN_MAX_COMPLETION_TOKENS must be greater than RETAIN_CHUNK_SIZE
         # to ensure the LLM has enough output capacity to extract facts from chunks
-        if self.retain_max_completion_tokens <= self.retain_chunk_size:
+        # (not applicable when provider is "none" since no LLM calls are made)
+        if self.llm_provider != "none" and self.retain_max_completion_tokens <= self.retain_chunk_size:
             raise ValueError(
                 f"Invalid configuration: HINDSIGHT_API_RETAIN_MAX_COMPLETION_TOKENS "
                 f"({self.retain_max_completion_tokens}) must be greater than "
@@ -957,6 +1169,7 @@ class HindsightConfig:
         config = cls(
             # Database
             database_url=os.getenv(ENV_DATABASE_URL, DEFAULT_DATABASE_URL),
+            migration_database_url=os.getenv(ENV_MIGRATION_DATABASE_URL) or None,
             database_schema=os.getenv(ENV_DATABASE_SCHEMA, DEFAULT_DATABASE_SCHEMA),
             vector_extension=os.getenv(ENV_VECTOR_EXTENSION, DEFAULT_VECTOR_EXTENSION).lower(),
             text_search_extension=os.getenv(ENV_TEXT_SEARCH_EXTENSION, DEFAULT_TEXT_SEARCH_EXTENSION).lower(),
@@ -972,6 +1185,7 @@ class HindsightConfig:
             llm_timeout=float(os.getenv(ENV_LLM_TIMEOUT, str(DEFAULT_LLM_TIMEOUT))),
             llm_groq_service_tier=os.getenv(ENV_LLM_GROQ_SERVICE_TIER, DEFAULT_LLM_GROQ_SERVICE_TIER),
             llm_openai_service_tier=os.getenv(ENV_LLM_OPENAI_SERVICE_TIER, DEFAULT_LLM_OPENAI_SERVICE_TIER),
+            llm_extra_body=json.loads(os.getenv(ENV_LLM_EXTRA_BODY, "null")),
             # Vertex AI
             llm_vertexai_project_id=os.getenv(ENV_LLM_VERTEXAI_PROJECT_ID) or DEFAULT_LLM_VERTEXAI_PROJECT_ID,
             llm_vertexai_region=os.getenv(ENV_LLM_VERTEXAI_REGION, DEFAULT_LLM_VERTEXAI_REGION),
@@ -979,6 +1193,14 @@ class HindsightConfig:
             or DEFAULT_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY,
             # Gemini safety settings (JSON-encoded list of {category, threshold} dicts)
             llm_gemini_safety_settings=json.loads(os.getenv(ENV_LLM_GEMINI_SAFETY_SETTINGS, "null")),
+            # Built-in llama.cpp configuration
+            llamacpp_model_path=os.getenv(ENV_LLAMACPP_MODEL_PATH) or None,
+            llamacpp_gpu_layers=int(os.getenv(ENV_LLAMACPP_GPU_LAYERS, str(DEFAULT_LLAMACPP_GPU_LAYERS))),
+            llamacpp_context_size=int(os.getenv(ENV_LLAMACPP_CONTEXT_SIZE, str(DEFAULT_LLAMACPP_CONTEXT_SIZE))),
+            llamacpp_chat_format=os.getenv(ENV_LLAMACPP_CHAT_FORMAT) or DEFAULT_LLAMACPP_CHAT_FORMAT,
+            llamacpp_no_grammar=os.getenv(ENV_LLAMACPP_NO_GRAMMAR, str(DEFAULT_LLAMACPP_NO_GRAMMAR)).lower()
+            in ("true", "1"),
+            llamacpp_extra_args=os.getenv(ENV_LLAMACPP_EXTRA_ARGS) or DEFAULT_LLAMACPP_EXTRA_ARGS,
             # Per-operation LLM config (None = use default)
             retain_llm_provider=os.getenv(ENV_RETAIN_LLM_PROVIDER) or None,
             retain_llm_api_key=os.getenv(ENV_RETAIN_LLM_API_KEY) or None,
@@ -1067,6 +1289,11 @@ class HindsightConfig:
             embeddings_cohere_api_key=os.getenv(ENV_EMBEDDINGS_COHERE_API_KEY) or os.getenv(ENV_COHERE_API_KEY),
             embeddings_cohere_model=os.getenv(ENV_EMBEDDINGS_COHERE_MODEL, DEFAULT_EMBEDDINGS_COHERE_MODEL),
             embeddings_cohere_base_url=os.getenv(ENV_EMBEDDINGS_COHERE_BASE_URL) or None,
+            # OpenRouter embeddings (with fallback to shared OpenRouter key, then LLM key)
+            embeddings_openrouter_api_key=os.getenv(ENV_EMBEDDINGS_OPENROUTER_API_KEY)
+            or os.getenv(ENV_OPENROUTER_API_KEY)
+            or os.getenv(ENV_LLM_API_KEY),
+            embeddings_openrouter_model=os.getenv(ENV_EMBEDDINGS_OPENROUTER_MODEL, DEFAULT_EMBEDDINGS_OPENROUTER_MODEL),
             # LiteLLM embeddings (with backward-compatible fallback to shared config)
             embeddings_litellm_api_base=os.getenv(ENV_EMBEDDINGS_LITELLM_API_BASE)
             or os.getenv(ENV_LITELLM_API_BASE, DEFAULT_LITELLM_API_BASE),
@@ -1078,6 +1305,26 @@ class HindsightConfig:
                 ENV_EMBEDDINGS_LITELLM_SDK_MODEL, DEFAULT_EMBEDDINGS_LITELLM_SDK_MODEL
             ),
             embeddings_litellm_sdk_api_base=os.getenv(ENV_EMBEDDINGS_LITELLM_SDK_API_BASE) or None,
+            embeddings_litellm_sdk_output_dimensions=int(v)
+            if (v := os.getenv(ENV_EMBEDDINGS_LITELLM_SDK_OUTPUT_DIMENSIONS))
+            else None,
+            embeddings_litellm_sdk_encoding_format=os.getenv(
+                ENV_EMBEDDINGS_LITELLM_SDK_ENCODING_FORMAT, DEFAULT_EMBEDDINGS_LITELLM_SDK_ENCODING_FORMAT
+            ),
+            # Gemini/Vertex AI embeddings (with fallback to LLM keys)
+            embeddings_gemini_api_key=os.getenv(ENV_EMBEDDINGS_GEMINI_API_KEY) or os.getenv(ENV_LLM_API_KEY),
+            embeddings_gemini_model=os.getenv(ENV_EMBEDDINGS_GEMINI_MODEL, DEFAULT_EMBEDDINGS_GEMINI_MODEL),
+            embeddings_gemini_output_dimensionality=int(
+                os.getenv(
+                    ENV_EMBEDDINGS_GEMINI_OUTPUT_DIMENSIONALITY,
+                    str(DEFAULT_EMBEDDINGS_GEMINI_OUTPUT_DIMENSIONALITY),
+                )
+            ),
+            embeddings_vertexai_project_id=os.getenv(ENV_EMBEDDINGS_VERTEXAI_PROJECT_ID)
+            or os.getenv(ENV_LLM_VERTEXAI_PROJECT_ID),
+            embeddings_vertexai_region=os.getenv(ENV_EMBEDDINGS_VERTEXAI_REGION) or os.getenv(ENV_LLM_VERTEXAI_REGION),
+            embeddings_vertexai_service_account_key=os.getenv(ENV_EMBEDDINGS_VERTEXAI_SERVICE_ACCOUNT_KEY)
+            or os.getenv(ENV_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY),
             # Reranker
             reranker_provider=os.getenv(ENV_RERANKER_PROVIDER, DEFAULT_RERANKER_PROVIDER),
             reranker_local_model=os.getenv(ENV_RERANKER_LOCAL_MODEL, DEFAULT_RERANKER_LOCAL_MODEL),
@@ -1092,6 +1339,15 @@ class HindsightConfig:
                 ENV_RERANKER_LOCAL_TRUST_REMOTE_CODE, str(DEFAULT_RERANKER_LOCAL_TRUST_REMOTE_CODE)
             ).lower()
             in ("true", "1"),
+            reranker_local_fp16=os.getenv(ENV_RERANKER_LOCAL_FP16, str(DEFAULT_RERANKER_LOCAL_FP16)).lower()
+            in ("true", "1"),
+            reranker_local_bucket_batching=os.getenv(
+                ENV_RERANKER_LOCAL_BUCKET_BATCHING, str(DEFAULT_RERANKER_LOCAL_BUCKET_BATCHING)
+            ).lower()
+            in ("true", "1"),
+            reranker_local_batch_size=int(
+                os.getenv(ENV_RERANKER_LOCAL_BATCH_SIZE, str(DEFAULT_RERANKER_LOCAL_BATCH_SIZE))
+            ),
             reranker_tei_url=os.getenv(ENV_RERANKER_TEI_URL),
             reranker_tei_batch_size=int(os.getenv(ENV_RERANKER_TEI_BATCH_SIZE, str(DEFAULT_RERANKER_TEI_BATCH_SIZE))),
             reranker_tei_max_concurrent=int(
@@ -1102,6 +1358,11 @@ class HindsightConfig:
             reranker_cohere_api_key=os.getenv(ENV_RERANKER_COHERE_API_KEY) or os.getenv(ENV_COHERE_API_KEY),
             reranker_cohere_model=os.getenv(ENV_RERANKER_COHERE_MODEL, DEFAULT_RERANKER_COHERE_MODEL),
             reranker_cohere_base_url=os.getenv(ENV_RERANKER_COHERE_BASE_URL) or None,
+            # OpenRouter reranker (with fallback to shared OpenRouter key, then LLM key)
+            reranker_openrouter_api_key=os.getenv(ENV_RERANKER_OPENROUTER_API_KEY)
+            or os.getenv(ENV_OPENROUTER_API_KEY)
+            or os.getenv(ENV_LLM_API_KEY),
+            reranker_openrouter_model=os.getenv(ENV_RERANKER_OPENROUTER_MODEL, DEFAULT_RERANKER_OPENROUTER_MODEL),
             # LiteLLM reranker (with backward-compatible fallback to shared config)
             reranker_litellm_api_base=os.getenv(ENV_RERANKER_LITELLM_API_BASE)
             or os.getenv(ENV_LITELLM_API_BASE, DEFAULT_LITELLM_API_BASE),
@@ -1117,6 +1378,19 @@ class HindsightConfig:
             # ZeroEntropy reranker
             reranker_zeroentropy_api_key=os.getenv(ENV_RERANKER_ZEROENTROPY_API_KEY),
             reranker_zeroentropy_model=os.getenv(ENV_RERANKER_ZEROENTROPY_MODEL, DEFAULT_RERANKER_ZEROENTROPY_MODEL),
+            reranker_zeroentropy_base_url=os.getenv(ENV_RERANKER_ZEROENTROPY_BASE_URL) or None,
+            # SiliconFlow reranker (Cohere-compatible /rerank endpoint)
+            reranker_siliconflow_api_key=os.getenv(ENV_RERANKER_SILICONFLOW_API_KEY),
+            reranker_siliconflow_model=os.getenv(ENV_RERANKER_SILICONFLOW_MODEL, DEFAULT_RERANKER_SILICONFLOW_MODEL),
+            reranker_siliconflow_base_url=os.getenv(
+                ENV_RERANKER_SILICONFLOW_BASE_URL, DEFAULT_RERANKER_SILICONFLOW_BASE_URL
+            ),
+            # Google Discovery Engine reranker (with fallback to LLM Vertex AI keys)
+            reranker_google_model=os.getenv(ENV_RERANKER_GOOGLE_MODEL, DEFAULT_RERANKER_GOOGLE_MODEL),
+            reranker_google_project_id=os.getenv(ENV_RERANKER_GOOGLE_PROJECT_ID)
+            or os.getenv(ENV_LLM_VERTEXAI_PROJECT_ID),
+            reranker_google_service_account_key=os.getenv(ENV_RERANKER_GOOGLE_SERVICE_ACCOUNT_KEY)
+            or os.getenv(ENV_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY),
             # Server
             host=os.getenv(ENV_HOST, DEFAULT_HOST),
             port=int(os.getenv(ENV_PORT, DEFAULT_PORT)),
@@ -1127,11 +1401,12 @@ class HindsightConfig:
             mcp_enabled_tools=[t.strip() for t in os.getenv(ENV_MCP_ENABLED_TOOLS).split(",") if t.strip()]
             if os.getenv(ENV_MCP_ENABLED_TOOLS)
             else DEFAULT_MCP_ENABLED_TOOLS,
+            mcp_stateless=os.getenv(ENV_MCP_STATELESS, str(DEFAULT_MCP_STATELESS)).lower() == "true",
             enable_bank_config_api=os.getenv(ENV_ENABLE_BANK_CONFIG_API, str(DEFAULT_ENABLE_BANK_CONFIG_API)).lower()
             == "true",
+            default_bank_template=_parse_default_bank_template(os.getenv(ENV_DEFAULT_BANK_TEMPLATE)),
             # Recall
             graph_retriever=os.getenv(ENV_GRAPH_RETRIEVER, DEFAULT_GRAPH_RETRIEVER),
-            mpfp_top_k_neighbors=int(os.getenv(ENV_MPFP_TOP_K_NEIGHBORS, str(DEFAULT_MPFP_TOP_K_NEIGHBORS))),
             recall_max_concurrent=int(os.getenv(ENV_RECALL_MAX_CONCURRENT, str(DEFAULT_RECALL_MAX_CONCURRENT))),
             recall_connection_budget=int(
                 os.getenv(ENV_RECALL_CONNECTION_BUDGET, str(DEFAULT_RECALL_CONNECTION_BUDGET))
@@ -1140,6 +1415,10 @@ class HindsightConfig:
             mental_model_refresh_concurrency=int(
                 os.getenv(ENV_MENTAL_MODEL_REFRESH_CONCURRENCY, str(DEFAULT_MENTAL_MODEL_REFRESH_CONCURRENCY))
             ),
+            link_expansion_per_entity_limit=int(
+                os.getenv(ENV_LINK_EXPANSION_PER_ENTITY_LIMIT, str(DEFAULT_LINK_EXPANSION_PER_ENTITY_LIMIT))
+            ),
+            link_expansion_timeout=float(os.getenv(ENV_LINK_EXPANSION_TIMEOUT, str(DEFAULT_LINK_EXPANSION_TIMEOUT))),
             # Optimization flags
             skip_llm_verification=os.getenv(ENV_SKIP_LLM_VERIFICATION, "false").lower() == "true",
             lazy_reranker=os.getenv(ENV_LAZY_RERANKER, "false").lower() == "true",
@@ -1157,6 +1436,8 @@ class HindsightConfig:
             ),
             retain_mission=os.getenv(ENV_RETAIN_MISSION) or DEFAULT_RETAIN_MISSION,
             retain_custom_instructions=os.getenv(ENV_RETAIN_CUSTOM_INSTRUCTIONS) or DEFAULT_RETAIN_CUSTOM_INSTRUCTIONS,
+            retain_default_strategy=os.getenv(ENV_RETAIN_DEFAULT_STRATEGY) or DEFAULT_RETAIN_DEFAULT_STRATEGY,
+            retain_strategies=DEFAULT_RETAIN_STRATEGIES,
             retain_batch_tokens=int(os.getenv(ENV_RETAIN_BATCH_TOKENS, str(DEFAULT_RETAIN_BATCH_TOKENS))),
             retain_entity_lookup=os.getenv(ENV_RETAIN_ENTITY_LOOKUP, DEFAULT_RETAIN_ENTITY_LOOKUP),
             retain_batch_enabled=os.getenv(ENV_RETAIN_BATCH_ENABLED, str(DEFAULT_RETAIN_BATCH_ENABLED)).lower()
@@ -1164,6 +1445,7 @@ class HindsightConfig:
             retain_batch_poll_interval_seconds=int(
                 os.getenv(ENV_RETAIN_BATCH_POLL_INTERVAL_SECONDS, str(DEFAULT_RETAIN_BATCH_POLL_INTERVAL_SECONDS))
             ),
+            retain_chunk_batch_size=int(os.getenv(ENV_RETAIN_CHUNK_BATCH_SIZE, str(DEFAULT_RETAIN_CHUNK_BATCH_SIZE))),
             # File storage
             file_storage_type=os.getenv(ENV_FILE_STORAGE_TYPE, DEFAULT_FILE_STORAGE_TYPE),
             file_storage_s3_bucket=os.getenv(ENV_FILE_STORAGE_S3_BUCKET) or None,
@@ -1223,6 +1505,9 @@ class HindsightConfig:
                 )
             ),
             observations_mission=os.getenv(ENV_OBSERVATIONS_MISSION) or DEFAULT_OBSERVATIONS_MISSION,
+            max_observations_per_scope=int(
+                os.getenv(ENV_MAX_OBSERVATIONS_PER_SCOPE, str(DEFAULT_MAX_OBSERVATIONS_PER_SCOPE))
+            ),
             entity_labels=None,
             entities_allow_free_form=True,
             # Database migrations
@@ -1242,12 +1527,23 @@ class HindsightConfig:
             worker_consolidation_max_slots=int(
                 os.getenv(ENV_WORKER_CONSOLIDATION_MAX_SLOTS, str(DEFAULT_WORKER_CONSOLIDATION_MAX_SLOTS))
             ),
+            retain_max_concurrent=int(os.getenv(ENV_RETAIN_MAX_CONCURRENT, str(DEFAULT_RETAIN_MAX_CONCURRENT))),
             # Reflect agent settings
             reflect_max_iterations=int(os.getenv(ENV_REFLECT_MAX_ITERATIONS, str(DEFAULT_REFLECT_MAX_ITERATIONS))),
             reflect_max_context_tokens=int(
                 os.getenv(ENV_REFLECT_MAX_CONTEXT_TOKENS, str(DEFAULT_REFLECT_MAX_CONTEXT_TOKENS))
             ),
+            reflect_wall_timeout=int(os.getenv(ENV_REFLECT_WALL_TIMEOUT, str(DEFAULT_REFLECT_WALL_TIMEOUT))),
             reflect_mission=os.getenv(ENV_REFLECT_MISSION) or None,
+            reflect_source_facts_max_tokens=int(
+                os.getenv(ENV_REFLECT_SOURCE_FACTS_MAX_TOKENS, str(DEFAULT_REFLECT_SOURCE_FACTS_MAX_TOKENS))
+            ),
+            recall_include_chunks=os.getenv(ENV_RECALL_INCLUDE_CHUNKS, str(DEFAULT_RECALL_INCLUDE_CHUNKS)).lower()
+            in ("true", "1", "yes"),
+            recall_max_tokens=int(os.getenv(ENV_RECALL_MAX_TOKENS, str(DEFAULT_RECALL_MAX_TOKENS))),
+            recall_chunks_max_tokens=int(
+                os.getenv(ENV_RECALL_CHUNKS_MAX_TOKENS, str(DEFAULT_RECALL_CHUNKS_MAX_TOKENS))
+            ),
             # Disposition settings (None = fall back to DB value)
             disposition_skepticism=int(os.getenv(ENV_DISPOSITION_SKEPTICISM))
             if os.getenv(ENV_DISPOSITION_SKEPTICISM)
@@ -1265,6 +1561,16 @@ class HindsightConfig:
             otel_exporter_otlp_headers=os.getenv(ENV_OTEL_EXPORTER_OTLP_HEADERS) or None,
             otel_service_name=os.getenv(ENV_OTEL_SERVICE_NAME, DEFAULT_OTEL_SERVICE_NAME),
             otel_deployment_environment=os.getenv(ENV_OTEL_DEPLOYMENT_ENVIRONMENT, DEFAULT_OTEL_DEPLOYMENT_ENVIRONMENT),
+            metrics_include_bank_id=os.getenv(ENV_METRICS_INCLUDE_BANK_ID, str(DEFAULT_METRICS_INCLUDE_BANK_ID)).lower()
+            in ("true", "1", "yes"),
+            # Audit log configuration (static, server-level only)
+            audit_log_enabled=os.getenv(ENV_AUDIT_LOG_ENABLED, str(DEFAULT_AUDIT_LOG_ENABLED)).lower() == "true",
+            audit_log_actions=[
+                a.strip() for a in os.getenv(ENV_AUDIT_LOG_ACTIONS, DEFAULT_AUDIT_LOG_ACTIONS).split(",") if a.strip()
+            ],
+            audit_log_retention_days=int(
+                os.getenv(ENV_AUDIT_LOG_RETENTION_DAYS, str(DEFAULT_AUDIT_LOG_RETENTION_DAYS))
+            ),
             # Webhook configuration (static, server-level only)
             webhook_url=os.getenv(ENV_WEBHOOK_URL) or DEFAULT_WEBHOOK_URL,
             webhook_secret=os.getenv(ENV_WEBHOOK_SECRET) or DEFAULT_WEBHOOK_SECRET,
@@ -1334,9 +1640,14 @@ class HindsightConfig:
 
         root_logger.addHandler(handler)
 
+        # Silence noisy third-party loggers
+        logging.getLogger("google_genai.models").setLevel(logging.WARNING)
+
     def log_config(self) -> None:
         """Log the current configuration (without sensitive values)."""
         logger.info(f"Database: {self.database_url} (schema: {self.database_schema})")
+        if self.migration_database_url:
+            logger.info(f"Migration database: {self.migration_database_url}")
         logger.info(f"LLM: provider={self.llm_provider}, model={self.llm_model}")
         if self.retain_llm_provider or self.retain_llm_model:
             retain_provider = self.retain_llm_provider or self.llm_provider

hindsight-api-slim/hindsight_api/config_resolver.py

@@ -10,7 +10,7 @@ multiple API servers.
 
 import json
 import logging
-from dataclasses import asdict
+from dataclasses import asdict, replace
 from typing import Any
 
 import asyncpg
@@ -239,6 +239,23 @@ class ConfigResolver:
                 logger.warning(f"Failed to check permissions for bank {bank_id}: {e}")
                 # Continue without permission check (fail open for backward compatibility)
 
+        # Validate entity_labels structure
+        if "entity_labels" in normalized_updates and normalized_updates["entity_labels"] is not None:
+            from .engine.retain.entity_labels import parse_entity_labels
+
+            try:
+                parse_entity_labels(normalized_updates["entity_labels"])
+            except Exception as e:
+                raise ValueError(f"Invalid entity_labels format: {e}")
+
+        # Validate retain_strategies: reject empty string keys
+        if "retain_strategies" in normalized_updates and normalized_updates["retain_strategies"]:
+            empty_keys = [k for k in normalized_updates["retain_strategies"] if not str(k).strip()]
+            if empty_keys:
+                raise ValueError(
+                    "Strategy names must not be empty strings. Remove entries with empty names before saving."
+                )
+
         # Merge with existing config (JSONB || operator)
         async with self.pool.acquire() as conn:
             await conn.execute(
@@ -273,3 +290,35 @@ class ConfigResolver:
             )
 
         logger.info(f"Reset bank config for {bank_id} to defaults")
+
+
+def apply_strategy(config: HindsightConfig, strategy_name: str) -> HindsightConfig:
+    """
+    Apply a named retain strategy's overrides on top of a resolved config.
+
+    A strategy is a named set of hierarchical field overrides stored in
+    config.retain_strategies. Any field in _HIERARCHICAL_FIELDS can be
+    overridden, including retain_extraction_mode, retain_chunk_size,
+    entity_labels, entities_allow_free_form, etc.
+
+    Unknown strategy names log a warning and return config unchanged.
+    Unknown or non-hierarchical fields in the strategy are silently ignored.
+    """
+    strategies = config.retain_strategies or {}
+    if strategy_name not in strategies:
+        logger.warning(f"Unknown retain strategy '{strategy_name}', using resolved config as-is")
+        return config
+
+    overrides = strategies[strategy_name]
+    if not isinstance(overrides, dict):
+        logger.warning(f"Retain strategy '{strategy_name}' is not a dict, skipping")
+        return config
+
+    configurable = HindsightConfig.get_configurable_fields()
+    filtered = {k: v for k, v in overrides.items() if k in configurable}
+
+    if not filtered:
+        return config
+
+    logger.debug(f"Applying retain strategy '{strategy_name}': {list(filtered.keys())}")
+    return replace(config, **filtered)

hindsight-api-slim/hindsight_api/engine/audit.py

@@ -0,0 +1,209 @@
+"""Audit logging for feature usage tracking.
+
+Provides fire-and-forget audit logging of all mutating and core operations
+(retain, recall, reflect, bank CRUD, etc.) across HTTP, MCP, and system transports.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import uuid
+from collections.abc import Callable
+from contextlib import asynccontextmanager
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from typing import Any
+
+import asyncpg
+
+from ..engine.db_utils import acquire_with_retry
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class AuditEntry:
+    """A single audit log entry."""
+
+    action: str
+    transport: str  # "http", "mcp", "system"
+    bank_id: str | None = None
+    started_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))
+    ended_at: datetime | None = None
+    request: dict[str, Any] | None = None
+    response: dict[str, Any] | None = None
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+
+def _json_default(obj: Any) -> str:
+    """JSON serializer for objects not serializable by default."""
+    if isinstance(obj, datetime):
+        return obj.isoformat()
+    if isinstance(obj, uuid.UUID):
+        return str(obj)
+    if isinstance(obj, bytes):
+        return "<bytes>"
+    if isinstance(obj, set):
+        return list(obj)
+    return str(obj)
+
+
+def _safe_json(data: Any) -> str | None:
+    """Serialize data to JSON string, returning None on failure."""
+    if data is None:
+        return None
+    try:
+        return json.dumps(data, default=_json_default)
+    except Exception:
+        logger.debug("Failed to serialize audit data", exc_info=True)
+        return None
+
+
+_SWEEP_INTERVAL_SECONDS = 3600  # Run retention sweep every hour
+
+
+class AuditLogger:
+    """Fire-and-forget audit log writer with optional retention sweep."""
+
+    def __init__(
+        self,
+        pool_getter: Callable[[], asyncpg.Pool | None],
+        schema_getter: Callable[[], str],
+        enabled: bool,
+        allowed_actions: list[str],
+        retention_days: int = -1,
+    ) -> None:
+        self._pool_getter = pool_getter
+        self._schema_getter = schema_getter
+        self._enabled = enabled
+        self._allowed_actions: frozenset[str] | None = frozenset(allowed_actions) if allowed_actions else None
+        self._retention_days = retention_days
+        self._sweep_task: asyncio.Task | None = None
+
+    def is_enabled(self, action: str) -> bool:
+        """Check if audit logging is enabled for this action."""
+        if not self._enabled:
+            return False
+        if self._allowed_actions is not None:
+            return action in self._allowed_actions
+        return True
+
+    def log_fire_and_forget(self, entry: AuditEntry) -> None:
+        """Schedule an audit write as a background task."""
+        if not self.is_enabled(entry.action):
+            return
+        try:
+            asyncio.create_task(self._safe_log(entry))
+        except RuntimeError:
+            # No running event loop (e.g. during shutdown)
+            logger.debug("Cannot schedule audit log write: no running event loop")
+
+    async def _safe_log(self, entry: AuditEntry) -> None:
+        """Write audit entry to DB. Errors are logged, never raised."""
+        pool = self._pool_getter()
+        if pool is None:
+            logger.debug("Audit log skipped: pool not available")
+            return
+        try:
+            schema = self._schema_getter()
+            table = f"{schema}.audit_log"
+            async with acquire_with_retry(pool, max_retries=1) as conn:
+                await conn.execute(
+                    f"""
+                    INSERT INTO {table}
+                        (id, action, transport, bank_id, started_at, ended_at, request, response, metadata)
+                    VALUES
+                        ($1, $2, $3, $4, $5, $6, $7::jsonb, $8::jsonb, $9::jsonb)
+                    """,
+                    uuid.uuid4(),
+                    entry.action,
+                    entry.transport,
+                    entry.bank_id,
+                    entry.started_at,
+                    entry.ended_at,
+                    _safe_json(entry.request),
+                    _safe_json(entry.response),
+                    _safe_json(entry.metadata) or "{}",
+                )
+        except Exception as e:
+            logger.warning(f"Audit log write failed for action={entry.action}: {e}")
+
+    def start_retention_sweep(self) -> None:
+        """Start the periodic retention sweep if retention is configured."""
+        if self._retention_days <= 0 or not self._enabled:
+            return
+        try:
+            self._sweep_task = asyncio.create_task(self._sweep_loop())
+        except RuntimeError:
+            logger.debug("Cannot start retention sweep: no running event loop")
+
+    async def stop_retention_sweep(self) -> None:
+        """Stop the periodic retention sweep."""
+        if self._sweep_task and not self._sweep_task.done():
+            self._sweep_task.cancel()
+            try:
+                await self._sweep_task
+            except asyncio.CancelledError:
+                pass
+            self._sweep_task = None
+
+    async def _sweep_loop(self) -> None:
+        """Periodically delete audit log entries older than retention_days."""
+        while True:
+            await self._run_sweep()
+            await asyncio.sleep(_SWEEP_INTERVAL_SECONDS)
+
+    async def _run_sweep(self) -> None:
+        """Delete expired audit log entries. Concurrent-safe via row-level deletes."""
+        pool = self._pool_getter()
+        if pool is None:
+            return
+        try:
+            schema = self._schema_getter()
+            table = f"{schema}.audit_log"
+            async with acquire_with_retry(pool, max_retries=1) as conn:
+                result = await conn.execute(
+                    f"DELETE FROM {table} WHERE started_at < NOW() - INTERVAL '{self._retention_days} days'"
+                )
+                if result and result != "DELETE 0":
+                    logger.info(f"Audit log retention sweep: {result}")
+        except Exception as e:
+            logger.warning(f"Audit log retention sweep failed: {e}")
+
+
+@asynccontextmanager
+async def audit_context(
+    audit_logger: AuditLogger | None,
+    action: str,
+    transport: str,
+    bank_id: str | None = None,
+    request: dict[str, Any] | None = None,
+    metadata: dict[str, Any] | None = None,
+):
+    """Async context manager that times the operation and writes audit on exit.
+
+    Usage:
+        async with audit_context(logger, "retain", "http", bank_id, request_dict) as entry:
+            result = await do_work()
+            entry.response = result_dict
+    """
+    if audit_logger is None or not audit_logger.is_enabled(action):
+        entry = AuditEntry(action=action, transport=transport, bank_id=bank_id)
+        yield entry
+        return
+
+    entry = AuditEntry(
+        action=action,
+        transport=transport,
+        bank_id=bank_id,
+        started_at=datetime.now(timezone.utc),
+        request=request,
+        metadata=metadata or {},
+    )
+    try:
+        yield entry
+    finally:
+        entry.ended_at = datetime.now(timezone.utc)
+        audit_logger.log_fire_and_forget(entry)

hindsight-api-slim/hindsight_api/engine/consolidation/consolidator.py

@@ -80,6 +80,7 @@ class _BatchLLMResult:
     deletes: list[_DeleteAction] = field(default_factory=list)
     obs_count: int = 0
     prompt_chars: int = 0
+    failed: bool = False
 
 
 @dataclass
@@ -118,6 +119,39 @@ def _aggregate_source_fields(source_mems: list[dict[str, Any]], tags: list[str]
     )
 
 
+async def _count_observations_for_scope(
+    conn: "Connection",
+    bank_id: str,
+    tags: list[str],
+) -> int:
+    """Count existing observations matching the given tag scope.
+
+    Returns the count of observations whose tags contain all specified tags.
+    Observations with no tags are not counted (the limit does not apply to them).
+    """
+    return await conn.fetchval(
+        f"SELECT COUNT(*) FROM {fq_table('memory_units')} "
+        f"WHERE bank_id = $1 AND fact_type = 'observation' AND tags @> $2::varchar[]",
+        bank_id,
+        tags,
+    )
+
+
+def _build_response_model(max_creates: int | None = None) -> type[_ConsolidationBatchResponse]:
+    """Build a response model, optionally constraining max creates via JSON schema."""
+    if max_creates is None or max_creates < 0:
+        return _ConsolidationBatchResponse
+
+    from pydantic import Field as PydanticField
+
+    clamped = max(max_creates, 0)
+
+    class _ConstrainedConsolidationBatchResponse(_ConsolidationBatchResponse):
+        creates: list[_CreateAction] = PydanticField(default=[], max_length=clamped)
+
+    return _ConstrainedConsolidationBatchResponse
+
+
 class ConsolidationPerfLog:
     """Performance logging for consolidation operations."""
 
@@ -219,6 +253,7 @@ async def run_consolidation_job(
             FROM {fq_table("memory_units")}
             WHERE bank_id = $1
               AND consolidated_at IS NULL
+              AND consolidation_failed_at IS NULL
               AND fact_type IN ('experience', 'world')
             """,
             bank_id,
@@ -240,6 +275,7 @@ async def run_consolidation_job(
         "observations_deleted": 0,
         "actions_executed": 0,
         "skipped": 0,
+        "memories_failed": 0,
     }
 
     # Track all unique tags from consolidated memories for mental model refresh filtering
@@ -257,6 +293,7 @@ async def run_consolidation_job(
                 FROM {fq_table("memory_units")}
                 WHERE bank_id = $1
                   AND consolidated_at IS NULL
+                  AND consolidation_failed_at IS NULL
                   AND fact_type IN ('experience', 'world')
                 ORDER BY created_at ASC
                 LIMIT $2
@@ -298,94 +335,141 @@ async def run_consolidation_job(
                 if memory_tags:
                     consolidated_tags.update(memory_tags)
 
-            async with pool.acquire() as conn:
-                # Determine observation_scopes for this batch. All memories in a batch share
-                # the same tags (enforced by tag_groups), so we only check the first memory.
-                # asyncpg returns JSONB columns as raw JSON strings, so parse if needed.
-                _obs_raw = llm_batch[0].get("observation_scopes") if llm_batch else None
-                _obs_parsed = json.loads(_obs_raw) if isinstance(_obs_raw, str) else _obs_raw
+            # Process llm_batch with adaptive splitting: on LLM failure, halve the sub-batch
+            # and retry, down to batch_size=1. Only if a single-memory batch still fails is
+            # the memory marked with consolidation_failed_at and excluded from future runs
+            # until explicitly retried via the API.
+            all_results: list[dict[str, Any]] = []
+            all_deleted = 0
+            succeeded_ids: list[Any] = []
+            failed_ids: list[Any] = []
 
-                # Resolve the scope spec into a concrete list[list[str]] (or None for combined).
-                if _obs_parsed == "per_tag":
-                    _memory_tags = llm_batch[0].get("tags") or []
-                    obs_tags_list = [[tag] for tag in _memory_tags] if _memory_tags else None
-                elif _obs_parsed == "all_combinations":
-                    _memory_tags = llm_batch[0].get("tags") or []
-                    obs_tags_list = (
-                        [
-                            list(combo)
-                            for r in range(1, len(_memory_tags) + 1)
-                            for combo in combinations(_memory_tags, r)
-                        ]
-                        if _memory_tags
-                        else None
-                    )
-                elif _obs_parsed == "combined" or _obs_parsed is None:
-                    obs_tags_list = None  # single combined pass (default behaviour)
-                else:
-                    # explicit list[list[str]]
-                    obs_tags_list = _obs_parsed
+            pending: list[list[dict[str, Any]]] = [llm_batch]
+            while pending:
+                sub_batch = pending.pop(0)
 
-                batch_deleted: int = 0
-                if obs_tags_list:
-                    # Multi-pass: run one observation consolidation pass per tag set
-                    results = []
-                    for obs_tags in obs_tags_list:
-                        pass_results, pass_deleted = await _process_memory_batch(
+                async with pool.acquire() as conn:
+                    # Determine observation_scopes for this sub-batch. All memories share
+                    # the same tags (enforced by tag_groups), so we only check the first memory.
+                    # asyncpg returns JSONB columns as raw JSON strings, so parse if needed.
+                    _obs_raw = sub_batch[0].get("observation_scopes") if sub_batch else None
+                    _obs_parsed = json.loads(_obs_raw) if isinstance(_obs_raw, str) else _obs_raw
+
+                    # Resolve the scope spec into a concrete list[list[str]] (or None for combined).
+                    if _obs_parsed == "per_tag":
+                        _memory_tags = sub_batch[0].get("tags") or []
+                        obs_tags_list = [[tag] for tag in _memory_tags] if _memory_tags else None
+                    elif _obs_parsed == "all_combinations":
+                        _memory_tags = sub_batch[0].get("tags") or []
+                        obs_tags_list = (
+                            [
+                                list(combo)
+                                for r in range(1, len(_memory_tags) + 1)
+                                for combo in combinations(_memory_tags, r)
+                            ]
+                            if _memory_tags
+                            else None
+                        )
+                    elif _obs_parsed == "combined" or _obs_parsed is None:
+                        obs_tags_list = None  # single combined pass (default behaviour)
+                    else:
+                        # explicit list[list[str]]
+                        obs_tags_list = _obs_parsed
+
+                    sub_deleted: int = 0
+                    sub_llm_failed = False
+                    if obs_tags_list:
+                        # Multi-pass: run one observation consolidation pass per tag set
+                        sub_results: list[dict[str, Any]] = []
+                        for obs_tags in obs_tags_list:
+                            pass_results, pass_deleted, pass_failed = await _process_memory_batch(
+                                conn=conn,
+                                memory_engine=memory_engine,
+                                llm_config=llm_config,
+                                bank_id=bank_id,
+                                memories=sub_batch,
+                                request_context=request_context,
+                                perf=perf,
+                                config=config,
+                                obs_tags_override=obs_tags,
+                            )
+                            sub_deleted += pass_deleted
+                            sub_llm_failed = sub_llm_failed or pass_failed
+                            # Merge results: prefer non-skipped actions
+                            if not sub_results:
+                                sub_results = pass_results
+                            else:
+                                for i, (existing, new) in enumerate(zip(sub_results, pass_results)):
+                                    if existing.get("action") == "skipped" and new.get("action") != "skipped":
+                                        sub_results[i] = new
+                                    elif existing.get("action") != "skipped" and new.get("action") != "skipped":
+                                        # Both did something — combine into "multiple"
+                                        existing_created = existing.get(
+                                            "created", 1 if existing.get("action") == "created" else 0
+                                        )
+                                        existing_updated = existing.get(
+                                            "updated", 1 if existing.get("action") == "updated" else 0
+                                        )
+                                        new_created = new.get("created", 1 if new.get("action") == "created" else 0)
+                                        new_updated = new.get("updated", 1 if new.get("action") == "updated" else 0)
+                                        total = existing_created + existing_updated + new_created + new_updated
+                                        sub_results[i] = {
+                                            "action": "multiple",
+                                            "created": existing_created + new_created,
+                                            "updated": existing_updated + new_updated,
+                                            "merged": 0,
+                                            "total_actions": total,
+                                        }
+                    else:
+                        # Normal single pass using the memory's own tags
+                        sub_results, sub_deleted, sub_llm_failed = await _process_memory_batch(
                             conn=conn,
                             memory_engine=memory_engine,
                             llm_config=llm_config,
                             bank_id=bank_id,
-                            memories=llm_batch,
+                            memories=sub_batch,
                             request_context=request_context,
                             perf=perf,
                             config=config,
-                            obs_tags_override=obs_tags,
                         )
-                        batch_deleted += pass_deleted
-                        # Merge results: prefer non-skipped actions
-                        if not results:
-                            results = pass_results
-                        else:
-                            for i, (existing, new) in enumerate(zip(results, pass_results)):
-                                if existing.get("action") == "skipped" and new.get("action") != "skipped":
-                                    results[i] = new
-                                elif existing.get("action") != "skipped" and new.get("action") != "skipped":
-                                    # Both did something — combine into "multiple"
-                                    existing_created = existing.get(
-                                        "created", 1 if existing.get("action") == "created" else 0
-                                    )
-                                    existing_updated = existing.get(
-                                        "updated", 1 if existing.get("action") == "updated" else 0
-                                    )
-                                    new_created = new.get("created", 1 if new.get("action") == "created" else 0)
-                                    new_updated = new.get("updated", 1 if new.get("action") == "updated" else 0)
-                                    total = existing_created + existing_updated + new_created + new_updated
-                                    results[i] = {
-                                        "action": "multiple",
-                                        "created": existing_created + new_created,
-                                        "updated": existing_updated + new_updated,
-                                        "merged": 0,
-                                        "total_actions": total,
-                                    }
-                else:
-                    # Normal single pass using the memory's own tags
-                    results, batch_deleted = await _process_memory_batch(
-                        conn=conn,
-                        memory_engine=memory_engine,
-                        llm_config=llm_config,
-                        bank_id=bank_id,
-                        memories=llm_batch,
-                        request_context=request_context,
-                        perf=perf,
-                        config=config,
-                    )
-                stats["observations_deleted"] += batch_deleted
 
-                await conn.executemany(
-                    f"UPDATE {fq_table('memory_units')} SET consolidated_at = NOW() WHERE id = $1",
-                    [(m["id"],) for m in llm_batch],
-                )
+                all_deleted += sub_deleted
+
+                if sub_llm_failed and len(sub_batch) > 1:
+                    # Split and retry with smaller batches
+                    mid = len(sub_batch) // 2
+                    logger.warning(
+                        f"[CONSOLIDATION] bank={bank_id} LLM failed for sub-batch of {len(sub_batch)},"
+                        f" splitting into {mid}/{len(sub_batch) - mid}"
+                    )
+                    pending[0:0] = [sub_batch[:mid], sub_batch[mid:]]
+                elif sub_llm_failed:
+                    # batch_size=1 and still failing — mark as permanently failed for now
+                    failed_ids.append(sub_batch[0]["id"])
+                    all_results.append({"action": "failed"})
+                    logger.warning(
+                        f"[CONSOLIDATION] bank={bank_id} LLM failed for single memory"
+                        f" {sub_batch[0]['id']}, marking consolidation_failed_at"
+                    )
+                else:
+                    succeeded_ids.extend(m["id"] for m in sub_batch)
+                    all_results.extend(sub_results)
+
+            # Commit consolidated_at / consolidation_failed_at in a single DB round-trip
+            async with pool.acquire() as conn:
+                if succeeded_ids:
+                    await conn.executemany(
+                        f"UPDATE {fq_table('memory_units')} SET consolidated_at = NOW() WHERE id = $1",
+                        [(mem_id,) for mem_id in succeeded_ids],
+                    )
+                if failed_ids:
+                    await conn.executemany(
+                        f"UPDATE {fq_table('memory_units')} SET consolidation_failed_at = NOW() WHERE id = $1",
+                        [(mem_id,) for mem_id in failed_ids],
+                    )
+
+            stats["observations_deleted"] += all_deleted
+            results = all_results
 
             # Checkpoint: abort if the operation (and thus the bank) was deleted mid-run.
             if operation_id and not await memory_engine._check_op_alive(operation_id):
@@ -413,6 +497,8 @@ async def run_consolidation_job(
                     stats["actions_executed"] += result.get("total_actions", 0)
                 elif action == "skipped":
                     stats["skipped"] += 1
+                elif action == "failed":
+                    stats["memories_failed"] += 1
 
             # Per-LLM-batch log
             llm_batch_time = time.time() - llm_batch_start
@@ -425,6 +511,7 @@ async def run_consolidation_job(
             batch_created = stats["observations_created"] - snap_stats["observations_created"]
             batch_updated = stats["observations_updated"] - snap_stats["observations_updated"]
             batch_skipped = stats["skipped"] - snap_stats["skipped"]
+            batch_failed = stats["memories_failed"] - snap_stats["memories_failed"]
             llm_calls_made = perf.llm_calls - snap_llm_calls
             logger.info(
                 f"[CONSOLIDATION] bank={bank_id} llm_batch #{llm_batch_num}"
@@ -432,7 +519,8 @@ async def run_consolidation_job(
                 f" | {stats['memories_processed']}/{total_count} processed"
                 f" | {', '.join(timing_parts)}"
                 f" | created={batch_created} updated={batch_updated} skipped={batch_skipped}"
-                f" | input_tokens=~{input_tokens}"
+                + (f" failed={batch_failed}" if batch_failed else "")
+                + f" | input_tokens=~{input_tokens}"
                 f" | avg={llm_batch_time / len(llm_batch):.3f}s/memory"
             )
 
@@ -584,7 +672,7 @@ async def _process_memory_batch(
     perf: ConsolidationPerfLog | None = None,
     config: Any = None,
     obs_tags_override: list[str] | None = None,
-) -> tuple[list[dict[str, Any]], int]:
+) -> tuple[list[dict[str, Any]], int, bool]:
     """
     Process a batch of memories in a single LLM call.
 
@@ -643,24 +731,6 @@ async def _process_memory_batch(
         if recall_result.source_facts:
             union_source_facts.update(recall_result.source_facts)
 
-    # 3. Single LLM call
-    t0 = time.time()
-    llm_result = await _consolidate_batch_with_llm(
-        llm_config=llm_config,
-        memories=memories,
-        union_observations=union_observations,
-        union_source_facts=union_source_facts,
-        config=config,
-    )
-    if perf:
-        perf.record_timing("llm", time.time() - t0)
-        perf.record_llm_call(llm_result.obs_count, llm_result.prompt_chars)
-
-    # 4. Sequential execution of creates / updates / deletes
-    # Track which memory indices participated so we can build per-memory results for stats
-    per_memory_created: set[str] = set()
-    per_memory_updated: set[str] = set()
-
     # Determine effective tag scope for observations.
     # When obs_tags_override is set, use it; otherwise use the memory's own tags.
     if obs_tags_override is not None:
@@ -669,28 +739,52 @@ async def _process_memory_batch(
         # All memories in the batch share the same tag set (enforced by batching)
         fact_tags = memories[0].get("tags") or [] if memories else []
 
+    # 2b. Compute remaining observation slots for this scope (if limit configured)
+    max_obs = config.max_observations_per_scope if config is not None else -1
+    remaining_observation_slots: int | None = None
+    if max_obs > 0 and fact_tags:
+        current_count = await _count_observations_for_scope(conn, bank_id, fact_tags)
+        remaining_observation_slots = max(max_obs - current_count, 0)
+        if remaining_observation_slots == 0:
+            logger.info(
+                f"[CONSOLIDATION] bank={bank_id} scope={fact_tags} at observation limit "
+                f"({current_count}/{max_obs}), only updates/deletes allowed"
+            )
+
+    # 3. Single LLM call
+    t0 = time.time()
+    llm_result = await _consolidate_batch_with_llm(
+        llm_config=llm_config,
+        memories=memories,
+        union_observations=union_observations,
+        union_source_facts=union_source_facts,
+        config=config,
+        remaining_observation_slots=remaining_observation_slots,
+        max_observations_per_scope=max_obs,
+    )
+    if perf:
+        perf.record_timing("llm", time.time() - t0)
+        perf.record_llm_call(llm_result.obs_count, llm_result.prompt_chars)
+
+    # 4. Sequential execution of deletes / updates / creates
+    # Deletes run first to free observation slots before creates consume them.
+    # Track which memory indices participated so we can build per-memory results for stats
+    per_memory_created: set[str] = set()
+    per_memory_updated: set[str] = set()
+
     mem_by_id = {str(m["id"]): m for m in memories}
 
-    for create in llm_result.creates:
-        source_mems = [mem_by_id[fid] for fid in create.source_fact_ids if fid in mem_by_id]
-        if not source_mems:
+    # Execute deletes first to free observation slots before creates consume them
+    deleted_count = 0
+    for delete in llm_result.deletes:
+        # Security: the observation must be present in the unioned recall
+        if not any(str(obs.id) == delete.observation_id for obs in union_observations):
+            logger.debug(
+                f"Batch consolidation: rejected delete — observation {delete.observation_id} not in unioned recall"
+            )
             continue
-        agg = _aggregate_source_fields(source_mems, tags=fact_tags)
-        await _execute_create_action(
-            conn=conn,
-            memory_engine=memory_engine,
-            bank_id=bank_id,
-            source_memory_ids=[m["id"] for m in source_mems],
-            text=create.text,
-            source_fact_tags=agg.tags,
-            event_date=agg.event_date,
-            occurred_start=agg.occurred_start,
-            occurred_end=agg.occurred_end,
-            mentioned_at=agg.mentioned_at,
-            perf=perf,
-        )
-        for m in source_mems:
-            per_memory_created.add(str(m["id"]))
+        await _execute_delete_action(conn=conn, bank_id=bank_id, observation_id=delete.observation_id)
+        deleted_count += 1
 
     for update in llm_result.updates:
         source_mems = [mem_by_id[fid] for fid in update.source_fact_ids if fid in mem_by_id]
@@ -721,16 +815,26 @@ async def _process_memory_batch(
         for m in source_mems:
             per_memory_updated.add(str(m["id"]))
 
-    deleted_count = 0
-    for delete in llm_result.deletes:
-        # Security: the observation must be present in the unioned recall
-        if not any(str(obs.id) == delete.observation_id for obs in union_observations):
-            logger.debug(
-                f"Batch consolidation: rejected delete — observation {delete.observation_id} not in unioned recall"
-            )
+    for create in llm_result.creates:
+        source_mems = [mem_by_id[fid] for fid in create.source_fact_ids if fid in mem_by_id]
+        if not source_mems:
             continue
-        await _execute_delete_action(conn=conn, bank_id=bank_id, observation_id=delete.observation_id)
-        deleted_count += 1
+        agg = _aggregate_source_fields(source_mems, tags=fact_tags)
+        await _execute_create_action(
+            conn=conn,
+            memory_engine=memory_engine,
+            bank_id=bank_id,
+            source_memory_ids=[m["id"] for m in source_mems],
+            text=create.text,
+            source_fact_tags=agg.tags,
+            event_date=agg.event_date,
+            occurred_start=agg.occurred_start,
+            occurred_end=agg.occurred_end,
+            mentioned_at=agg.mentioned_at,
+            perf=perf,
+        )
+        for m in source_mems:
+            per_memory_created.add(str(m["id"]))
 
     # Build per-memory result dicts for the stats tracker in the outer loop
     results: list[dict[str, Any]] = []
@@ -747,7 +851,7 @@ async def _process_memory_batch(
         else:
             results.append({"action": "skipped", "reason": "no_durable_knowledge"})
 
-    return results, deleted_count
+    return results, deleted_count, llm_result.failed
 
 
 def _min_date(dates: "Any") -> "datetime | None":
@@ -1028,6 +1132,8 @@ async def _consolidate_batch_with_llm(
     union_observations: "list[MemoryFact]",
     union_source_facts: "dict[str, MemoryFact]",
     config: Any = None,
+    remaining_observation_slots: int | None = None,
+    max_observations_per_scope: int = -1,
 ) -> _BatchLLMResult:
     """Single LLM call for a batch of facts against a pooled set of observations."""
     if union_observations:
@@ -1051,24 +1157,51 @@ async def _consolidate_batch_with_llm(
 
     facts_lines = "\n".join(_fact_line(m) for m in memories)
 
+    # Build capacity note for the prompt when observation limit is configured
+    observation_capacity_note: str | None = None
+    if remaining_observation_slots is not None and max_observations_per_scope > 0:
+        if remaining_observation_slots == 0:
+            observation_capacity_note = (
+                f"OBSERVATION LIMIT REACHED ({max_observations_per_scope}/{max_observations_per_scope}). "
+                "Only UPDATE or DELETE existing observations. Do NOT create new ones — "
+                "merge new knowledge into existing observations via UPDATE."
+            )
+        elif remaining_observation_slots <= len(memories):
+            observation_capacity_note = (
+                f"This scope has {remaining_observation_slots} observation slot(s) remaining "
+                f"(out of {max_observations_per_scope}). Prefer UPDATE over CREATE when possible."
+            )
+
     observations_mission = config.observations_mission if config is not None else None
-    prompt_template = build_batch_consolidation_prompt(observations_mission)
+    prompt_template = build_batch_consolidation_prompt(observations_mission, observation_capacity_note)
     prompt = prompt_template.format(
         facts_text=facts_lines,
         observations_text=observations_text,
     )
 
+    # Use a constrained response model when observation limit is active
+    response_model = _build_response_model(max_creates=remaining_observation_slots)
+
     max_attempts = 3
     last_exc: Exception | None = None
     for attempt in range(1, max_attempts + 1):
         try:
             response: _ConsolidationBatchResponse = await llm_config.call(
                 messages=[{"role": "user", "content": prompt}],
-                response_format=_ConsolidationBatchResponse,
+                response_format=response_model,
                 scope="consolidation",
             )
+            # Defensive truncation: some LLM providers may not enforce JSON schema max_length
+            creates = response.creates
+            if remaining_observation_slots is not None and remaining_observation_slots >= 0:
+                if len(creates) > remaining_observation_slots:
+                    logger.info(
+                        f"[CONSOLIDATION] Truncating {len(creates)} creates to {remaining_observation_slots} "
+                        f"(max_observations_per_scope={max_observations_per_scope})"
+                    )
+                    creates = creates[:remaining_observation_slots]
             return _BatchLLMResult(
-                creates=response.creates,
+                creates=creates,
                 updates=response.updates,
                 deletes=response.deletes,
                 obs_count=len(union_observations),
@@ -1081,7 +1214,7 @@ async def _consolidate_batch_with_llm(
     logger.error(
         f"[CONSOLIDATION] LLM batch call failed after {max_attempts} attempts, skipping batch. Last error: {last_exc}"
     )
-    return _BatchLLMResult(obs_count=len(union_observations), prompt_chars=len(prompt))
+    return _BatchLLMResult(obs_count=len(union_observations), prompt_chars=len(prompt), failed=True)
 
 
 async def _create_observation_directly(

hindsight-api-slim/hindsight_api/engine/consolidation/prompts.py

@@ -5,10 +5,24 @@ _DEFAULT_MISSION = "Track every detail: names, numbers, dates, places, and relat
 
 # Processing rules — always present regardless of mission
 _PROCESSING_RULES = """Processing rules (always apply):
-- REDUNDANT: same info worded differently → UPDATE the existing observation.
-- CONTRADICTION/UPDATE: capture both states with temporal markers ("used to X, now Y").
-- RESOLVE REFERENCES: when a new fact provides a concrete value resolving a vague placeholder in an existing observation (e.g. "home country", "hometown", "birthplace", "native language", "her ex", "that city"), UPDATE the observation to embed the resolved value explicitly. Example: new fact says "grandma in Sweden" + existing observation says "moved from her home country" → update to "home country is Sweden".
-- NEVER merge observations about different people or unrelated topics."""
+
+1. ONE OBSERVATION PER DISTINCT FACET: each observation tracks exactly one specific facet — a count ("has 3 items"), a named entity ("has a dog named Rex"), a relationship ("works at Google"), etc. Never merge different facets into one observation.
+
+2. MATCH BY ENTITY/FACET, NOT TOPIC: when deciding whether to UPDATE vs CREATE, match on the specific entity or facet. "Sold item X" updates only the X observation. "Now has 5 items" updates only the count observation. Do not update observations about different entities just because they share a general topic.
+
+3. STATE CHANGES — UPDATE CONCISELY: when a fact changes the state of something ("sold X", "X died", "moved to Y"), UPDATE the matching observation to reflect the current state. Include dates when available. Keep it concise — only information about THAT specific facet. Example: "User owned a dog named Rex who died on March 15, 2025". Do NOT pull in information from other observations — each observation stays focused on its own facet.
+
+4. CASCADE TO ALL AFFECTED OBSERVATIONS: a state change may affect multiple observations. For example, if entity C is removed from a group, update BOTH the individual observation for C AND any list/group observation that includes C (remove C from the list while keeping all other members intact).
+
+5. NO COMPUTATION: you do not have the full picture — never calculate, derive, or adjust numeric values. If the user says "I have 2 dogs" and then "I have a dog named Rex", do NOT update the count to 3 — you don't know if Rex is one of the 2 or a new one. If the user says "I sold X", do NOT decrement a count. Only update a count when the user explicitly states a new count. Synthesize and consolidate what was stated, but never do arithmetic or logical deductions.
+
+6. SAME FACET → UPDATE, NOT CREATE: a new count supersedes the old count — UPDATE the existing count observation, don't create a second one. If there's an existing observation for the same specific facet, always UPDATE it rather than creating a duplicate.
+
+7. PRESERVE HISTORY: observations that record significant events (sold, died, moved, changed) are important history — never DELETE them. Only delete an observation when it is restated identically or truly meaningless. Be very conservative with deletes.
+
+8. RESOLVE REFERENCES: when a new fact provides a concrete value for a vague placeholder in an existing observation (e.g., "home country" → "Sweden"), UPDATE to embed the resolved value.
+
+9. NEVER merge observations about different people or unrelated topics."""
 
 # Data section — format placeholders {facts_text} and {observations_text} are substituted at call time
 _BATCH_DATA_SECTION = """
@@ -26,8 +40,8 @@ Each observation includes:
 - source_memories: array of supporting facts with their text and dates
 
 Compare the facts against existing observations:
-- Same topic as an existing observation → UPDATE it (observation_id + source_fact_ids)
-- New topic with durable knowledge → CREATE a new observation (source_fact_ids)
+- Same facet as an existing observation → UPDATE it (observation_id + source_fact_ids)
+- New facet with durable knowledge → CREATE a new observation (source_fact_ids)
 - Cross-reference facts within the batch: a later fact may resolve a vague reference in an earlier one
 - Purely ephemeral facts → omit them unless the MISSION above explicitly targets such data (e.g. timestamped events, session state, screen content)"""
 
@@ -66,7 +80,10 @@ Rules:
 - Return {{"creates": [], "updates": [], "deletes": []}} if nothing durable is found."""
 
 
-def build_batch_consolidation_prompt(observations_mission: str | None = None) -> str:
+def build_batch_consolidation_prompt(
+    observations_mission: str | None = None,
+    observation_capacity_note: str | None = None,
+) -> str:
     """
     Build the consolidation prompt for batch mode (multiple facts per LLM call).
 
@@ -75,9 +92,13 @@ def build_batch_consolidation_prompt(observations_mission: str | None = None) ->
     """
     mission = observations_mission or _DEFAULT_MISSION
 
+    capacity_section = ""
+    if observation_capacity_note:
+        capacity_section = f"\n\n## CAPACITY CONSTRAINT\n{observation_capacity_note}"
+
     return (
         "You are a memory consolidation system. Synthesize facts into observations "
         "and merge with existing observations when appropriate.\n\n"
-        f"## MISSION\n{mission}\n\n"
+        f"## MISSION\n{mission}{capacity_section}\n\n"
         f"{_PROCESSING_RULES}" + _BATCH_DATA_SECTION + _BATCH_OUTPUT_FORMAT
     )

hindsight-api-slim/hindsight_api/engine/cross_encoder.py

@@ -20,14 +20,18 @@ from ..config import (
     DEFAULT_RERANKER_COHERE_MODEL,
     DEFAULT_RERANKER_FLASHRANK_CACHE_DIR,
     DEFAULT_RERANKER_FLASHRANK_MODEL,
+    DEFAULT_RERANKER_GOOGLE_MODEL,
     DEFAULT_RERANKER_LITELLM_MAX_TOKENS_PER_DOC,
     DEFAULT_RERANKER_LITELLM_MODEL,
     DEFAULT_RERANKER_LITELLM_SDK_MODEL,
+    DEFAULT_RERANKER_LOCAL_BATCH_SIZE,
     DEFAULT_RERANKER_LOCAL_FORCE_CPU,
     DEFAULT_RERANKER_LOCAL_MAX_CONCURRENT,
     DEFAULT_RERANKER_LOCAL_MODEL,
     DEFAULT_RERANKER_LOCAL_TRUST_REMOTE_CODE,
     DEFAULT_RERANKER_PROVIDER,
+    DEFAULT_RERANKER_SILICONFLOW_BASE_URL,
+    DEFAULT_RERANKER_SILICONFLOW_MODEL,
     DEFAULT_RERANKER_TEI_BATCH_SIZE,
     DEFAULT_RERANKER_TEI_MAX_CONCURRENT,
     DEFAULT_RERANKER_ZEROENTROPY_MODEL,
@@ -35,12 +39,14 @@ from ..config import (
     ENV_RERANKER_COHERE_MODEL,
     ENV_RERANKER_FLASHRANK_CACHE_DIR,
     ENV_RERANKER_FLASHRANK_MODEL,
+    ENV_RERANKER_GOOGLE_PROJECT_ID,
     ENV_RERANKER_LITELLM_SDK_API_KEY,
     ENV_RERANKER_LOCAL_FORCE_CPU,
     ENV_RERANKER_LOCAL_MAX_CONCURRENT,
     ENV_RERANKER_LOCAL_MODEL,
     ENV_RERANKER_LOCAL_TRUST_REMOTE_CODE,
     ENV_RERANKER_PROVIDER,
+    ENV_RERANKER_SILICONFLOW_API_KEY,
     ENV_RERANKER_TEI_BATCH_SIZE,
     ENV_RERANKER_TEI_MAX_CONCURRENT,
     ENV_RERANKER_TEI_URL,
@@ -111,6 +117,9 @@ class LocalSTCrossEncoder(CrossEncoderModel):
         max_concurrent: int = 4,
         force_cpu: bool = False,
         trust_remote_code: bool = False,
+        fp16: bool = False,
+        bucket_batching: bool = False,
+        batch_size: int = DEFAULT_RERANKER_LOCAL_BATCH_SIZE,
     ):
         """
         Initialize local SentenceTransformers cross-encoder.
@@ -125,10 +134,20 @@ class LocalSTCrossEncoder(CrossEncoderModel):
             trust_remote_code: Allow loading models with custom code (security risk).
                               Required for some models like jina-reranker-v2-base-multilingual.
                               Default: False (disabled for security)
+            fp16: Use FP16 (half precision) inference. Faster on MPS and CUDA,
+                  may be slower on CPU. Default: False (opt-in via env var).
+            bucket_batching: Sort pairs by token length before batching to reduce
+                            padding waste. 36-54% speedup, quality-identical.
+                            Default: False (opt-in via env var).
+            batch_size: Batch size for predict() calls. Optimal values vary by
+                       hardware and model (MPS: 32, CUDA: 128+). Default: 32.
         """
         self.model_name = model_name or DEFAULT_RERANKER_LOCAL_MODEL
         self.force_cpu = force_cpu
         self.trust_remote_code = trust_remote_code
+        self.fp16 = fp16
+        self.bucket_batching = bucket_batching
+        self.batch_size = batch_size
         self._model = None
         LocalSTCrossEncoder._max_concurrent = max_concurrent
 
@@ -176,6 +195,24 @@ class LocalSTCrossEncoder(CrossEncoderModel):
             except Exception as e:
                 logger.warning(f"Failed to detect GPU/MPS, falling back to CPU: {e}")
 
+        # Patch transformers 5.x compatibility for models using XLM-RoBERTa
+        # (e.g., jina-reranker-v2-base-multilingual). transformers 5.x removed
+        # create_position_ids_from_input_ids as a module-level function; the custom
+        # code in these models still references it. This monkey-patch restores it.
+        try:
+            import transformers.models.xlm_roberta.modeling_xlm_roberta as xlm_module
+            from transformers.models.xlm_roberta.modeling_xlm_roberta import XLMRobertaEmbeddings
+
+            if not hasattr(xlm_module, "create_position_ids_from_input_ids"):
+                setattr(
+                    xlm_module,
+                    "create_position_ids_from_input_ids",
+                    XLMRobertaEmbeddings.create_position_ids_from_input_ids,
+                )
+                logger.info("Reranker: applied transformers 5.x compatibility patch for XLM-RoBERTa")
+        except Exception:
+            pass
+
         # Suppress verbose transformers warnings during model loading
         # This suppresses the "UNEXPECTED" warnings from CrossEncoder which are harmless
         # but look alarming to users (e.g., "embeddings.position_ids | UNEXPECTED")
@@ -200,6 +237,12 @@ class LocalSTCrossEncoder(CrossEncoderModel):
                 # Restore original logging level
                 transformers_logger.setLevel(original_level)
 
+        # FP16 inference: convert model weights to half precision.
+        # Empirically validated: 27-36% faster on MPS, quality-identical (20/20 overlap).
+        if self.fp16 and device != "cpu":
+            self._model.model.half()
+            logger.info("Reranker: FP16 inference enabled")
+
         # Initialize shared executor (limited workers naturally limits concurrency)
         if LocalSTCrossEncoder._executor is None:
             LocalSTCrossEncoder._executor = ThreadPoolExecutor(
@@ -211,8 +254,32 @@ class LocalSTCrossEncoder(CrossEncoderModel):
             logger.info("Reranker: local provider initialized (using existing executor)")
 
     def _predict_sync(self, pairs: list[tuple[str, str]]) -> list[float]:
-        """Synchronous prediction wrapper for thread pool execution."""
-        scores = self._model.predict(pairs, show_progress_bar=False)
+        """Synchronous prediction wrapper for thread pool execution.
+
+        Supports two optimizations (controlled via .env):
+        - bucket_batching: sort pairs by token length to reduce padding waste (36-54% speedup)
+        - batch_size: explicit batch size for predict() calls (MPS optimal: 32)
+        """
+        import numpy as np
+
+        if self.bucket_batching and len(pairs) > 1:
+            # Sort pairs by approximate token length to create homogeneous batches.
+            # This eliminates padding waste — short pairs aren't padded to the length
+            # of the longest pair in the batch. Quality-identical by construction.
+            lengths = [len(pairs[i][0]) + len(pairs[i][1]) for i in range(len(pairs))]
+            sorted_indices = sorted(range(len(pairs)), key=lambda i: lengths[i])
+            sorted_pairs = [pairs[i] for i in sorted_indices]
+
+            sorted_scores = self._model.predict(sorted_pairs, batch_size=self.batch_size, show_progress_bar=False)
+            sorted_scores = sorted_scores.tolist() if hasattr(sorted_scores, "tolist") else list(sorted_scores)
+
+            # Restore original order
+            scores = [0.0] * len(pairs)
+            for new_pos, orig_idx in enumerate(sorted_indices):
+                scores[orig_idx] = sorted_scores[new_pos]
+            return scores
+
+        scores = self._model.predict(pairs, batch_size=self.batch_size, show_progress_bar=False)
         return scores.tolist() if hasattr(scores, "tolist") else list(scores)
 
     async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
@@ -454,6 +521,84 @@ class RemoteTEICrossEncoder(CrossEncoderModel):
         return await self._predict_async(pairs)
 
 
+class _CohereCompatibleRerankClient:
+    """
+    Internal HTTP client for Cohere-compatible /rerank endpoints.
+
+    Shared by all providers that speak the Cohere rerank wire format —
+    {model, query, documents[, top_n]} request and
+    {results: [{index, relevance_score}, ...]} response. This covers
+    SiliconFlow, ZeroEntropy, Jina, Voyage, BGE self-hosted, and Cohere
+    itself when reached via a custom base_url (e.g. Azure AI Foundry).
+
+    Not a CrossEncoderModel — providers compose it and expose their own
+    provider_name / initialization logging.
+    """
+
+    def __init__(
+        self,
+        api_key: str,
+        model: str,
+        rerank_url: str,
+        timeout: float = 60.0,
+        include_top_n: bool = True,
+    ):
+        self.api_key = api_key
+        self.model = model
+        self.rerank_url = rerank_url
+        self.timeout = timeout
+        self.include_top_n = include_top_n
+        self._async_client: httpx.AsyncClient | None = None
+
+    async def initialize(self) -> None:
+        if self._async_client is not None:
+            return
+        self._async_client = httpx.AsyncClient(
+            timeout=self.timeout,
+            headers={
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json",
+            },
+        )
+
+    async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
+        if self._async_client is None:
+            raise RuntimeError("Reranker not initialized. Call initialize() first.")
+
+        if not pairs:
+            return []
+
+        query_groups: dict[str, list[tuple[int, str]]] = {}
+        for idx, (query, text) in enumerate(pairs):
+            query_groups.setdefault(query, []).append((idx, text))
+
+        all_scores = [0.0] * len(pairs)
+
+        for query, indexed_texts in query_groups.items():
+            texts = [text for _, text in indexed_texts]
+            indices = [idx for idx, _ in indexed_texts]
+
+            body: dict[str, object] = {
+                "model": self.model,
+                "query": query,
+                "documents": texts,
+                "return_documents": False,
+            }
+            if self.include_top_n:
+                body["top_n"] = len(texts)
+
+            response = await self._async_client.post(self.rerank_url, json=body)
+            response.raise_for_status()
+            result = response.json()
+
+            for item in result.get("results", []):
+                original_idx = item["index"]
+                score = item["relevance_score"]
+                all_scores[indices[original_idx]] = score
+
+        return all_scores
+
+
 class CohereCrossEncoder(CrossEncoderModel):
     """
     Cohere cross-encoder implementation using the Cohere Rerank API.
@@ -482,6 +627,20 @@ class CohereCrossEncoder(CrossEncoderModel):
         self.base_url = base_url
         self.timeout = timeout
         self._client = None
+        # Used when base_url is set (Azure AI Foundry and other Cohere-compatible hosts).
+        # Azure endpoints already include the full invoke path, so rerank_url == base_url
+        # and top_n is omitted to match the existing Azure contract.
+        self._http_client: _CohereCompatibleRerankClient | None = (
+            _CohereCompatibleRerankClient(
+                api_key=api_key,
+                model=model,
+                rerank_url=base_url,
+                timeout=timeout,
+                include_top_n=False,
+            )
+            if base_url
+            else None
+        )
 
     @property
     def provider_name(self) -> str:
@@ -489,23 +648,24 @@ class CohereCrossEncoder(CrossEncoderModel):
 
     async def initialize(self) -> None:
         """Initialize the Cohere client."""
-        if self._client is not None:
+        if self._client is not None or (self._http_client and self._http_client._async_client):
             return
 
-        try:
-            import cohere
-        except ImportError:
-            raise ImportError("cohere is required for CohereCrossEncoder. Install it with: pip install cohere")
-
         base_url_msg = f" at {self.base_url}" if self.base_url else ""
         logger.info(f"Reranker: initializing Cohere provider with model {self.model}{base_url_msg}")
 
-        # Build client kwargs, only including base_url if set (for Azure or custom endpoints)
-        client_kwargs = {"api_key": self.api_key, "timeout": self.timeout}
-        if self.base_url:
-            client_kwargs["base_url"] = self.base_url
-        self._client = cohere.Client(**client_kwargs)
-        logger.info("Reranker: Cohere provider initialized")
+        if self._http_client is not None:
+            await self._http_client.initialize()
+            logger.info("Reranker: Cohere provider initialized (Cohere-compatible HTTP endpoint)")
+        else:
+            # For native Cohere API, use the official SDK
+            try:
+                import cohere
+            except ImportError:
+                raise ImportError("cohere is required for CohereCrossEncoder. Install it with: pip install cohere")
+
+            self._client = cohere.Client(api_key=self.api_key, timeout=self.timeout)
+            logger.info("Reranker: Cohere provider initialized")
 
     async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
         """
@@ -517,25 +677,24 @@ class CohereCrossEncoder(CrossEncoderModel):
         Returns:
             List of relevance scores
         """
-        if self._client is None:
+        if self._client is None and self._http_client is None:
             raise RuntimeError("Reranker not initialized. Call initialize() first.")
 
         if not pairs:
             return []
 
-        # Run sync Cohere API calls in thread pool
-        loop = asyncio.get_event_loop()
-        return await loop.run_in_executor(None, self._predict_sync, pairs)
+        if self._http_client is not None:
+            return await self._http_client.predict(pairs)
 
-    def _predict_sync(self, pairs: list[tuple[str, str]]) -> list[float]:
-        """Synchronous predict implementation for Cohere API."""
-        # Group pairs by query for efficient batching
-        # Cohere rerank expects one query with multiple documents
+        # Run sync Cohere SDK calls in thread pool
+        loop = asyncio.get_event_loop()
+        return await loop.run_in_executor(None, self._predict_sync_sdk, pairs)
+
+    def _predict_sync_sdk(self, pairs: list[tuple[str, str]]) -> list[float]:
+        """Synchronous predict using the native Cohere SDK."""
         query_groups: dict[str, list[tuple[int, str]]] = {}
         for idx, (query, text) in enumerate(pairs):
-            if query not in query_groups:
-                query_groups[query] = []
-            query_groups[query].append((idx, text))
+            query_groups.setdefault(query, []).append((idx, text))
 
         all_scores = [0.0] * len(pairs)
 
@@ -550,7 +709,6 @@ class CohereCrossEncoder(CrossEncoderModel):
                 return_documents=False,
             )
 
-            # Map scores back to original positions
             for result in response.results:
                 original_idx = result.index
                 score = result.relevance_score
@@ -567,94 +725,80 @@ class ZeroEntropyCrossEncoder(CrossEncoderModel):
     See: https://docs.zeroentropy.dev/models
     """
 
-    RERANK_URL = "https://api.zeroentropy.dev/v1/models/rerank"
+    DEFAULT_BASE_URL = "https://api.zeroentropy.dev"
+    RERANK_PATH = "/v1/models/rerank"
 
     def __init__(
         self,
         api_key: str,
         model: str = DEFAULT_RERANKER_ZEROENTROPY_MODEL,
+        base_url: str | None = None,
         timeout: float = 60.0,
     ):
-        """
-        Initialize ZeroEntropy cross-encoder client.
-
-        Args:
-            api_key: ZeroEntropy API key
-            model: ZeroEntropy rerank model name (default: zerank-2)
-            timeout: Request timeout in seconds (default: 60.0)
-        """
-        self.api_key = api_key
         self.model = model
-        self.timeout = timeout
-        self._async_client: httpx.AsyncClient | None = None
+        self.base_url = base_url.rstrip("/") if base_url else self.DEFAULT_BASE_URL
+        self._client = _CohereCompatibleRerankClient(
+            api_key=api_key,
+            model=model,
+            rerank_url=f"{self.base_url}{self.RERANK_PATH}",
+            timeout=timeout,
+        )
 
     @property
     def provider_name(self) -> str:
         return "zeroentropy"
 
     async def initialize(self) -> None:
-        """Initialize the async HTTP client."""
-        if self._async_client is not None:
+        if self._client._async_client is not None:
             return
-
         logger.info(f"Reranker: initializing ZeroEntropy provider with model {self.model}")
-        self._async_client = httpx.AsyncClient(
-            timeout=self.timeout,
-            headers={
-                "Authorization": f"Bearer {self.api_key}",
-                "Content-Type": "application/json",
-            },
-        )
+        await self._client.initialize()
         logger.info("Reranker: ZeroEntropy provider initialized")
 
     async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
-        """
-        Score query-document pairs using the ZeroEntropy Rerank API.
+        return await self._client.predict(pairs)
 
-        Args:
-            pairs: List of (query, document) tuples to score
 
-        Returns:
-            List of relevance scores
-        """
-        if self._async_client is None:
-            raise RuntimeError("Reranker not initialized. Call initialize() first.")
+class SiliconFlowCrossEncoder(CrossEncoderModel):
+    """
+    SiliconFlow cross-encoder implementation.
 
-        if not pairs:
-            return []
+    SiliconFlow (https://siliconflow.cn) exposes a Cohere-compatible /rerank
+    endpoint. Shares the HTTP client with ZeroEntropy/Cohere-custom-endpoint
+    via _CohereCompatibleRerankClient.
+    """
 
-        # Group pairs by query for efficient batching
-        query_groups: dict[str, list[tuple[int, str]]] = {}
-        for idx, (query, text) in enumerate(pairs):
-            if query not in query_groups:
-                query_groups[query] = []
-            query_groups[query].append((idx, text))
+    RERANK_PATH = "/rerank"
 
-        all_scores = [0.0] * len(pairs)
+    def __init__(
+        self,
+        api_key: str,
+        model: str = DEFAULT_RERANKER_SILICONFLOW_MODEL,
+        base_url: str = DEFAULT_RERANKER_SILICONFLOW_BASE_URL,
+        timeout: float = 60.0,
+    ):
+        self.model = model
+        self.base_url = base_url.rstrip("/")
+        self._client = _CohereCompatibleRerankClient(
+            api_key=api_key,
+            model=model,
+            rerank_url=f"{self.base_url}{self.RERANK_PATH}",
+            timeout=timeout,
+        )
 
-        for query, indexed_texts in query_groups.items():
-            texts = [text for _, text in indexed_texts]
-            indices = [idx for idx, _ in indexed_texts]
+    @property
+    def provider_name(self) -> str:
+        return "siliconflow"
 
-            response = await self._async_client.post(
-                self.RERANK_URL,
-                json={
-                    "model": self.model,
-                    "query": query,
-                    "documents": texts,
-                    "top_n": len(texts),
-                },
-            )
-            response.raise_for_status()
-            result = response.json()
+    async def initialize(self) -> None:
+        if self._client._async_client is not None:
+            return
+        logger.info(f"Reranker: initializing SiliconFlow provider at {self.base_url} with model {self.model}")
+        await self._client.initialize()
+        logger.info("Reranker: SiliconFlow provider initialized")
 
-            # Map scores back to original positions
-            for item in result.get("results", []):
-                original_idx = item["index"]
-                score = item["relevance_score"]
-                all_scores[indices[original_idx]] = score
-
-        return all_scores
+    async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
+        return await self._client.predict(pairs)
 
 
 class RRFPassthroughCrossEncoder(CrossEncoderModel):
@@ -1106,14 +1250,31 @@ class JinaMLXCrossEncoder(CrossEncoderModel):
         if self._reranker is not None:
             return
 
+        # Pre-warm transformers.AutoTokenizer to fully populate the transformers
+        # namespace before mlx_lm imports it. transformers 5.x uses _LazyModule,
+        # which has an unguarded window where `from transformers import AutoTokenizer`
+        # raises ImportError if another thread is concurrently initializing the
+        # namespace (e.g. embeddings init in an executor thread).
+        # See: https://github.com/vectorize-io/hindsight/issues/994
+        import transformers
+
+        _ = transformers.AutoTokenizer
+
         try:
             import mlx.core  # noqa: F401
             import mlx_lm  # noqa: F401
-        except ImportError:
+        except ImportError as exc:
+            # Only swallow "package not installed" errors. Anything else (e.g. a
+            # transitive import failure inside mlx_lm) must surface verbatim so
+            # the real cause is debuggable instead of being masked by a generic
+            # "install mlx" message.
+            msg = str(exc)
+            if "mlx" not in msg and "mlx_lm" not in msg:
+                raise
             raise ImportError(
                 "mlx and mlx-lm are required for JinaMLXCrossEncoder. "
                 "Install with: pip install mlx>=0.31.0 mlx-lm>=0.31.1 safetensors>=0.6.2"
-            )
+            ) from exc
 
         loop = asyncio.get_event_loop()
         await loop.run_in_executor(None, self._load_model)
@@ -1167,6 +1328,164 @@ class JinaMLXCrossEncoder(CrossEncoderModel):
         return await loop.run_in_executor(None, self._predict_sync, pairs)
 
 
+class GoogleCrossEncoder(CrossEncoderModel):
+    """
+    Google Discovery Engine cross-encoder using the Ranking REST API.
+
+    Uses httpx + google-auth for lightweight REST calls (no gRPC/protobuf).
+    Supports ADC (Application Default Credentials) or service account key file.
+
+    Available models:
+    - semantic-ranker-default-004: Best quality, 1024 tokens/record (recommended)
+    - semantic-ranker-fast-004: Lower latency, 1024 tokens/record
+
+    Max 200 records per API request. Location is always "global".
+    """
+
+    MAX_RECORDS_PER_REQUEST = 200
+    API_BASE = "https://discoveryengine.googleapis.com/v1"
+    SCOPES = ["https://www.googleapis.com/auth/cloud-platform"]
+
+    def __init__(
+        self,
+        project_id: str,
+        model: str = DEFAULT_RERANKER_GOOGLE_MODEL,
+        service_account_key: str | None = None,
+        location: str = "global",
+        timeout: float = 60.0,
+    ):
+        """
+        Initialize Google Discovery Engine cross-encoder.
+
+        Args:
+            project_id: Google Cloud project ID
+            model: Ranking model name (default: semantic-ranker-default-004)
+            service_account_key: Path to service account JSON key file.
+                                If None, uses Application Default Credentials (ADC).
+            location: API location (default: "global")
+            timeout: Request timeout in seconds (default: 60.0)
+        """
+        self.project_id = project_id
+        self.model = model
+        self.service_account_key = service_account_key
+        self.location = location
+        self.timeout = timeout
+        self._credentials = None
+        self._client: httpx.Client | None = None
+        self._rank_url: str | None = None
+
+    @property
+    def provider_name(self) -> str:
+        return "google"
+
+    def _get_auth_headers(self) -> dict[str, str]:
+        """Get Authorization header with a fresh access token."""
+        import google.auth.transport.requests
+
+        if not self._credentials.valid:
+            self._credentials.refresh(google.auth.transport.requests.Request())
+        return {"Authorization": f"Bearer {self._credentials.token}"}
+
+    async def initialize(self) -> None:
+        """Initialize credentials and HTTP client."""
+        if self._client is not None:
+            return
+
+        auth_method = "ADC" if not self.service_account_key else "service_account"
+        logger.info(
+            f"Reranker: initializing Google Discovery Engine provider "
+            f"(project={self.project_id}, model={self.model}, auth={auth_method})"
+        )
+        if self.service_account_key:
+            try:
+                from google.oauth2 import service_account
+            except ImportError:
+                raise ImportError(
+                    "google-auth is required for GoogleCrossEncoder. Install it with: pip install google-auth"
+                )
+            self._credentials = service_account.Credentials.from_service_account_file(
+                self.service_account_key,
+                scopes=self.SCOPES,
+            )
+        else:
+            try:
+                import google.auth
+            except ImportError:
+                raise ImportError(
+                    "google-auth is required for GoogleCrossEncoder. Install it with: pip install google-auth"
+                )
+            self._credentials, _ = google.auth.default(scopes=self.SCOPES)
+
+        ranking_config = f"projects/{self.project_id}/locations/{self.location}/rankingConfigs/default_ranking_config"
+        self._rank_url = f"{self.API_BASE}/{ranking_config}:rank"
+        self._client = httpx.Client(timeout=self.timeout)
+
+        logger.info("Reranker: Google Discovery Engine provider initialized")
+
+    def _predict_sync(self, pairs: list[tuple[str, str]]) -> list[float]:
+        """Synchronous predict via REST API."""
+        if not pairs:
+            return []
+
+        # Group pairs by query
+        query_groups: dict[str, list[tuple[int, str]]] = {}
+        for idx, (query, text) in enumerate(pairs):
+            if query not in query_groups:
+                query_groups[query] = []
+            query_groups[query].append((idx, text))
+
+        all_scores = [0.0] * len(pairs)
+
+        for query, indexed_texts in query_groups.items():
+            texts = [text for _, text in indexed_texts]
+            indices = [idx for idx, _ in indexed_texts]
+
+            # Process in batches of MAX_RECORDS_PER_REQUEST
+            for batch_start in range(0, len(texts), self.MAX_RECORDS_PER_REQUEST):
+                batch_texts = texts[batch_start : batch_start + self.MAX_RECORDS_PER_REQUEST]
+                batch_indices = indices[batch_start : batch_start + self.MAX_RECORDS_PER_REQUEST]
+
+                records = [{"id": str(i), "content": text} for i, text in enumerate(batch_texts)]
+
+                response = self._client.post(
+                    self._rank_url,
+                    headers=self._get_auth_headers(),
+                    json={
+                        "model": self.model,
+                        "query": query,
+                        "records": records,
+                        "topN": len(records),
+                    },
+                )
+                response.raise_for_status()
+                result = response.json()
+
+                for record in result.get("records", []):
+                    local_idx = int(record["id"])
+                    all_scores[batch_indices[local_idx]] = record["score"]
+
+        return all_scores
+
+    async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
+        """
+        Score query-document pairs using Google Discovery Engine Ranking API.
+
+        Args:
+            pairs: List of (query, document) tuples to score
+
+        Returns:
+            List of relevance scores (0-1, higher = more relevant)
+        """
+        if self._client is None:
+            raise RuntimeError("Reranker not initialized. Call initialize() first.")
+
+        if not pairs:
+            return []
+
+        loop = asyncio.get_event_loop()
+        return await loop.run_in_executor(None, self._predict_sync, pairs)
+
+
 def create_cross_encoder_from_env() -> CrossEncoderModel:
     """
     Create a CrossEncoderModel instance based on configuration.
@@ -1196,6 +1515,9 @@ def create_cross_encoder_from_env() -> CrossEncoderModel:
             max_concurrent=config.reranker_local_max_concurrent,
             force_cpu=config.reranker_local_force_cpu,
             trust_remote_code=config.reranker_local_trust_remote_code,
+            fp16=config.reranker_local_fp16,
+            bucket_batching=config.reranker_local_bucket_batching,
+            batch_size=config.reranker_local_batch_size,
         )
     elif provider == "cohere":
         api_key = config.reranker_cohere_api_key
@@ -1206,6 +1528,18 @@ def create_cross_encoder_from_env() -> CrossEncoderModel:
             model=config.reranker_cohere_model,
             base_url=config.reranker_cohere_base_url,
         )
+    elif provider == "openrouter":
+        api_key = config.reranker_openrouter_api_key
+        if not api_key:
+            raise ValueError(
+                "HINDSIGHT_API_RERANKER_OPENROUTER_API_KEY, HINDSIGHT_API_OPENROUTER_API_KEY, "
+                f"or HINDSIGHT_API_LLM_API_KEY is required when {ENV_RERANKER_PROVIDER} is 'openrouter'"
+            )
+        return CohereCrossEncoder(
+            api_key=api_key,
+            model=config.reranker_openrouter_model,
+            base_url="https://openrouter.ai/api/v1/rerank",
+        )
     elif provider == "flashrank":
         model = os.environ.get(ENV_RERANKER_FLASHRANK_MODEL, DEFAULT_RERANKER_FLASHRANK_MODEL)
         cache_dir = os.environ.get(ENV_RERANKER_FLASHRANK_CACHE_DIR, DEFAULT_RERANKER_FLASHRANK_CACHE_DIR)
@@ -1239,11 +1573,34 @@ def create_cross_encoder_from_env() -> CrossEncoderModel:
             api_key=api_key,
             model=config.reranker_zeroentropy_model,
         )
+    elif provider == "siliconflow":
+        api_key = config.reranker_siliconflow_api_key
+        if not api_key:
+            raise ValueError(
+                f"{ENV_RERANKER_SILICONFLOW_API_KEY} is required when {ENV_RERANKER_PROVIDER} is 'siliconflow'"
+            )
+        return SiliconFlowCrossEncoder(
+            api_key=api_key,
+            model=config.reranker_siliconflow_model,
+            base_url=config.reranker_siliconflow_base_url,
+        )
+    elif provider == "google":
+        project_id = config.reranker_google_project_id
+        if not project_id:
+            raise ValueError(
+                f"{ENV_RERANKER_GOOGLE_PROJECT_ID} (or HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID) "
+                f"is required when {ENV_RERANKER_PROVIDER} is 'google'"
+            )
+        return GoogleCrossEncoder(
+            project_id=project_id,
+            model=config.reranker_google_model,
+            service_account_key=config.reranker_google_service_account_key,
+        )
     elif provider == "rrf":
         return RRFPassthroughCrossEncoder()
     elif provider == "jina-mlx":
         return JinaMLXCrossEncoder()
     else:
         raise ValueError(
-            f"Unknown reranker provider: {provider}. Supported: 'local', 'tei', 'cohere', 'zeroentropy', 'flashrank', 'litellm', 'litellm-sdk', 'rrf', 'jina-mlx'"
+            f"Unknown reranker provider: {provider}. Supported: 'local', 'tei', 'cohere', 'zeroentropy', 'siliconflow', 'google', 'flashrank', 'litellm', 'litellm-sdk', 'rrf', 'jina-mlx'"
         )

hindsight-api-slim/hindsight_api/engine/embeddings.py

@@ -13,11 +13,13 @@ import logging
 import os
 import warnings
 from abc import ABC, abstractmethod
+from urllib.parse import parse_qs, urlparse, urlunparse
 
 import httpx
 
 from ..config import (
     DEFAULT_EMBEDDINGS_COHERE_MODEL,
+    DEFAULT_EMBEDDINGS_GEMINI_MODEL,
     DEFAULT_EMBEDDINGS_LITELLM_MODEL,
     DEFAULT_EMBEDDINGS_LITELLM_SDK_MODEL,
     DEFAULT_EMBEDDINGS_LOCAL_FORCE_CPU,
@@ -27,6 +29,7 @@ from ..config import (
     DEFAULT_EMBEDDINGS_PROVIDER,
     DEFAULT_LITELLM_API_BASE,
     ENV_EMBEDDINGS_COHERE_API_KEY,
+    ENV_EMBEDDINGS_GEMINI_API_KEY,
     ENV_EMBEDDINGS_LITELLM_SDK_API_KEY,
     ENV_EMBEDDINGS_LOCAL_FORCE_CPU,
     ENV_EMBEDDINGS_LOCAL_MODEL,
@@ -426,9 +429,19 @@ class OpenAIEmbeddings(Embeddings):
         logger.info(f"Embeddings: initializing OpenAI provider with model {self.model}{base_url_msg}")
 
         # Build client kwargs, only including base_url if set (for Azure or custom endpoints)
+        # Parse query parameters from base_url (e.g. ?api-version=xxx for Azure OpenAI)
+        # and pass them as default_query so they're included in every request.
         client_kwargs = {"api_key": self.api_key, "max_retries": self.max_retries}
         if self.base_url:
-            client_kwargs["base_url"] = self.base_url
+            parsed = urlparse(self.base_url)
+            if parsed.query:
+                clean_url = urlunparse(parsed._replace(query=""))
+                client_kwargs["base_url"] = clean_url
+                default_query = {k: v[0] for k, v in parse_qs(parsed.query).items()}
+                client_kwargs["default_query"] = default_query
+                self.base_url = clean_url
+            else:
+                client_kwargs["base_url"] = self.base_url
         self._client = OpenAI(**client_kwargs)
 
         # Try to get dimension from known models, otherwise do a test embedding
@@ -741,8 +754,10 @@ class LiteLLMSDKEmbeddings(Embeddings):
         api_key: str,
         model: str = DEFAULT_EMBEDDINGS_LITELLM_SDK_MODEL,
         api_base: str | None = None,
+        output_dimensions: int | None = None,
         batch_size: int = 100,
         timeout: float = 60.0,
+        encoding_format: str | None = "float",
     ):
         """
         Initialize LiteLLM SDK embeddings client.
@@ -751,14 +766,19 @@ class LiteLLMSDKEmbeddings(Embeddings):
             api_key: API key for the embedding provider
             model: Model name with provider prefix (e.g., "cohere/embed-english-v3.0")
             api_base: Custom base URL for API (optional)
+            output_dimensions: Optional output embedding dimensions (provider-dependent)
             batch_size: Maximum batch size for embedding requests (default: 100)
             timeout: Request timeout in seconds (default: 60.0)
+            encoding_format: Encoding format for embeddings (default: "float").
+                Set to None or empty string to omit (needed for Voyage AI, Gemini).
         """
         self.api_key = api_key
         self.model = model
         self.api_base = api_base
+        self.output_dimensions = output_dimensions
         self.batch_size = batch_size
         self.timeout = timeout
+        self.encoding_format = encoding_format or None
         self._litellm = None  # Will be set during initialization
         self._dimension: int | None = None
 
@@ -794,10 +814,13 @@ class LiteLLMSDKEmbeddings(Embeddings):
                 "model": self.model,
                 "input": ["test"],
                 "api_key": self.api_key,
-                "encoding_format": "float",
             }
+            if self.encoding_format:
+                embed_kwargs["encoding_format"] = self.encoding_format
             if self.api_base:
                 embed_kwargs["api_base"] = self.api_base
+            if self.output_dimensions is not None:
+                embed_kwargs["dimensions"] = self.output_dimensions
 
             # Use async embedding method (standard in litellm)
             response = await self._litellm.aembedding(**embed_kwargs)
@@ -841,10 +864,13 @@ class LiteLLMSDKEmbeddings(Embeddings):
                     "model": self.model,
                     "input": batch,
                     "api_key": self.api_key,
-                    "encoding_format": "float",
                 }
+                if self.encoding_format:
+                    embed_kwargs["encoding_format"] = self.encoding_format
                 if self.api_base:
                     embed_kwargs["api_base"] = self.api_base
+                if self.output_dimensions is not None:
+                    embed_kwargs["dimensions"] = self.output_dimensions
 
                 # Use sync embedding (litellm doesn't have async in thread-safe way)
                 response = self._litellm.embedding(**embed_kwargs)
@@ -866,6 +892,179 @@ class LiteLLMSDKEmbeddings(Embeddings):
         return all_embeddings
 
 
+class GeminiEmbeddings(Embeddings):
+    """
+    Google embeddings via the google.genai SDK.
+
+    Supports both:
+    1. Gemini API (api.generativeai.google.com) with API key authentication
+    2. Vertex AI with service account or Application Default Credentials (ADC)
+
+    Uses the embed_content API: client.models.embed_content(model, contents)
+    """
+
+    def __init__(
+        self,
+        model: str = DEFAULT_EMBEDDINGS_GEMINI_MODEL,
+        api_key: str | None = None,
+        vertexai_project_id: str | None = None,
+        vertexai_region: str | None = None,
+        vertexai_service_account_key: str | None = None,
+        output_dimensionality: int | None = None,
+        batch_size: int = 100,
+    ):
+        self.model = model
+        self.api_key = api_key
+        self.vertexai_project_id = vertexai_project_id
+        self.vertexai_region = vertexai_region or "us-central1"
+        self.vertexai_service_account_key = vertexai_service_account_key
+        self.output_dimensionality = output_dimensionality
+        self.batch_size = batch_size
+        self._client = None
+        self._dimension: int | None = None
+        self._is_vertexai = vertexai_project_id is not None
+        self._embed_config = None  # EmbedContentConfig, built during initialize()
+
+    @property
+    def provider_name(self) -> str:
+        return "google"
+
+    @property
+    def dimension(self) -> int:
+        if self._dimension is None:
+            raise RuntimeError("Embeddings not initialized. Call initialize() first.")
+        return self._dimension
+
+    async def initialize(self) -> None:
+        """Initialize the Google genai client and detect embedding dimension."""
+        if self._client is not None:
+            return
+
+        from google import genai
+        from google.genai import types as genai_types
+
+        if self._is_vertexai:
+            self._init_vertexai(genai)
+        else:
+            self._init_gemini(genai)
+
+        # Build EmbedContentConfig if output_dimensionality is set
+        if self.output_dimensionality is not None:
+            self._embed_config = genai_types.EmbedContentConfig(
+                output_dimensionality=self.output_dimensionality,
+            )
+
+        # Detect dimension via a test embedding (respects output_dimensionality)
+        embed_kwargs = {"model": self.model, "contents": ["test"]}
+        if self._embed_config is not None:
+            embed_kwargs["config"] = self._embed_config
+
+        result = self._client.models.embed_content(**embed_kwargs)  # type: ignore[union-attr]
+        if result.embeddings and len(result.embeddings) > 0:
+            self._dimension = len(result.embeddings[0].values)
+
+        auth_mode = "vertex_ai" if self._is_vertexai else "api_key"
+        logger.info(
+            f"Embeddings: google provider initialized (auth: {auth_mode}, model: {self.model}, dim: {self._dimension})"
+        )
+
+    def _init_gemini(self, genai) -> None:
+        """Initialize Gemini API client with API key."""
+        if not self.api_key:
+            raise ValueError("Gemini embeddings provider requires an API key")
+
+        self._client = genai.Client(api_key=self.api_key)
+        logger.info(f"Embeddings: initializing Gemini provider with model {self.model}")
+
+    def _init_vertexai(self, genai) -> None:
+        """Initialize Vertex AI client with project, region, and credentials."""
+        if not self.vertexai_project_id:
+            raise ValueError(
+                "HINDSIGHT_API_EMBEDDINGS_VERTEXAI_PROJECT_ID (or HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID) "
+                "is required for Vertex AI embeddings provider."
+            )
+
+        auth_method = "ADC"
+        credentials = None
+
+        if self.vertexai_service_account_key:
+            try:
+                from google.oauth2 import service_account
+            except ImportError:
+                raise ImportError(
+                    "Vertex AI service account auth requires 'google-auth' package. "
+                    "Install with: pip install google-auth"
+                )
+            credentials = service_account.Credentials.from_service_account_file(
+                self.vertexai_service_account_key,
+                scopes=["https://www.googleapis.com/auth/cloud-platform"],
+            )
+            auth_method = "service_account"
+            logger.info(f"Embeddings: Vertex AI using service account key: {self.vertexai_service_account_key}")
+
+        # Strip google/ prefix from model name — native SDK uses bare names
+        if self.model.startswith("google/"):
+            self.model = self.model[len("google/") :]
+
+        client_kwargs = {
+            "vertexai": True,
+            "project": self.vertexai_project_id,
+            "location": self.vertexai_region,
+        }
+        if credentials is not None:
+            client_kwargs["credentials"] = credentials
+
+        self._client = genai.Client(**client_kwargs)
+        logger.info(
+            f"Embeddings: initializing Vertex AI provider "
+            f"(project={self.vertexai_project_id}, region={self.vertexai_region}, "
+            f"model={self.model}, auth={auth_method})"
+        )
+
+    def encode(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings using the Google genai SDK.
+
+        Args:
+            texts: List of text strings to encode
+
+        Returns:
+            List of embedding vectors
+        """
+        if self._client is None:
+            raise RuntimeError("Embeddings not initialized. Call initialize() first.")
+
+        if not texts:
+            return []
+
+        all_embeddings = []
+
+        # Process in batches
+        for i in range(0, len(texts), self.batch_size):
+            batch = texts[i : i + self.batch_size]
+
+            embed_kwargs = {"model": self.model, "contents": batch}
+            if self._embed_config is not None:
+                embed_kwargs["config"] = self._embed_config
+
+            result = self._client.models.embed_content(**embed_kwargs)
+
+            all_embeddings.extend([emb.values for emb in result.embeddings])
+
+        # L2-normalize when output_dimensionality is set — Gemini only returns
+        # normalized vectors at full 3072 dims; truncated dims need re-normalization
+        # for accurate cosine similarity.
+        if self.output_dimensionality is not None:
+            import numpy as np
+
+            arr = np.array(all_embeddings)
+            norms = np.linalg.norm(arr, axis=1, keepdims=True)
+            norms[norms == 0] = 1
+            all_embeddings = (arr / norms).tolist()
+
+        return all_embeddings
+
+
 def create_embeddings_from_env() -> Embeddings:
     """
     Create an Embeddings instance based on configuration.
@@ -902,6 +1101,18 @@ def create_embeddings_from_env() -> Embeddings:
         model = os.environ.get(ENV_EMBEDDINGS_OPENAI_MODEL, DEFAULT_EMBEDDINGS_OPENAI_MODEL)
         base_url = os.environ.get(ENV_EMBEDDINGS_OPENAI_BASE_URL) or None
         return OpenAIEmbeddings(api_key=api_key, model=model, base_url=base_url)
+    elif provider == "openrouter":
+        api_key = config.embeddings_openrouter_api_key
+        if not api_key:
+            raise ValueError(
+                "HINDSIGHT_API_EMBEDDINGS_OPENROUTER_API_KEY, HINDSIGHT_API_OPENROUTER_API_KEY, "
+                f"or {ENV_LLM_API_KEY} is required when {ENV_EMBEDDINGS_PROVIDER} is 'openrouter'"
+            )
+        return OpenAIEmbeddings(
+            api_key=api_key,
+            model=config.embeddings_openrouter_model,
+            base_url="https://openrouter.ai/api/v1",
+        )
     elif provider == "cohere":
         api_key = config.embeddings_cohere_api_key
         if not api_key:
@@ -927,9 +1138,30 @@ def create_embeddings_from_env() -> Embeddings:
             api_key=api_key,
             model=config.embeddings_litellm_sdk_model,
             api_base=config.embeddings_litellm_sdk_api_base,
+            output_dimensions=config.embeddings_litellm_sdk_output_dimensions,
+            encoding_format=config.embeddings_litellm_sdk_encoding_format,
+        )
+    elif provider == "google":
+        vertexai_project_id = config.embeddings_vertexai_project_id
+        if vertexai_project_id:
+            api_key = None  # Vertex AI uses ADC or service account
+        else:
+            api_key = config.embeddings_gemini_api_key
+            if not api_key:
+                raise ValueError(
+                    f"{ENV_EMBEDDINGS_GEMINI_API_KEY} or {ENV_LLM_API_KEY} is required "
+                    f"when {ENV_EMBEDDINGS_PROVIDER} is 'google' (set VERTEXAI_PROJECT_ID for Vertex AI auth instead)"
+                )
+        return GeminiEmbeddings(
+            model=config.embeddings_gemini_model,
+            api_key=api_key,
+            vertexai_project_id=vertexai_project_id,
+            vertexai_region=config.embeddings_vertexai_region,
+            vertexai_service_account_key=config.embeddings_vertexai_service_account_key,
+            output_dimensionality=config.embeddings_gemini_output_dimensionality,
         )
     else:
         raise ValueError(
             f"Unknown embeddings provider: {provider}. "
-            f"Supported: 'local', 'tei', 'openai', 'cohere', 'litellm', 'litellm-sdk'"
+            f"Supported: 'local', 'tei', 'openai', 'cohere', 'google', 'litellm', 'litellm-sdk'"
         )

hindsight-api-slim/hindsight_api/engine/entity_resolver.py

@@ -75,6 +75,7 @@ class EntityResolver:
         """
         self.pool = pool
         self.entity_lookup = entity_lookup
+        self._pg_trgm_checked = False
         # Keyed by asyncio task id so concurrent retain batches never mix their
         # pending updates.  flush_pending_stats() pops only the calling task's items.
         self._pending_stats: dict[int, list[_EntityStat]] = {}
@@ -85,6 +86,19 @@ class EntityResolver:
         task = asyncio.current_task()
         return id(task) if task is not None else 0
 
+    def discard_pending_stats(self) -> None:
+        """
+        Discard accumulated entity stats and co-occurrence counts for the current task.
+
+        Call this on any exception path between resolve_entities_batch /
+        link_units_to_entities_batch and flush_pending_stats() to prevent the
+        per-task dicts from growing unbounded when tasks fail before flushing.
+        Safe to call even if no entries exist for the current task.
+        """
+        key = self._task_key()
+        self._pending_stats.pop(key, None)
+        self._pending_cooccurrences.pop(key, None)
+
     async def flush_pending_stats(self) -> None:
         """
         Flush accumulated entity stats and co-occurrence counts for the current task.
@@ -202,6 +216,20 @@ class EntityResolver:
         taxonomy_lookup: set[str] | None = None,
     ) -> list[str]:
         if self.entity_lookup == "trigram":
+            # Auto-detect pg_trgm availability on first call and fall back to
+            # "full" strategy if the extension is not installed.  See #626.
+            if not self._pg_trgm_checked:
+                self._pg_trgm_checked = True
+                has_trgm = await conn.fetchval("SELECT EXISTS(SELECT 1 FROM pg_extension WHERE extname = 'pg_trgm')")
+                if not has_trgm:
+                    logger.warning(
+                        "pg_trgm extension is not available — falling back to 'full' "
+                        "entity lookup strategy. Install pg_trgm for faster entity "
+                        "resolution on large banks. See: "
+                        "https://github.com/vectorize-io/hindsight/issues/626"
+                    )
+                    self.entity_lookup = "full"
+                    return await self._resolve_entities_batch_full(conn, bank_id, entities_data, unit_event_date)
             return await self._resolve_entities_batch_trigram(conn, bank_id, entities_data, unit_event_date)
         return await self._resolve_entities_batch_full(conn, bank_id, entities_data, unit_event_date)
 
@@ -289,8 +317,13 @@ class EntityResolver:
         entity_texts = list(set(e["text"] for e in entities_data))
 
         # Fetch candidates for all unique entity texts in a single batched query.
-        # The trigram % operator uses the GIN index; the substring conditions cover
-        # exact prefix/suffix matches that trigrams might miss at low similarity.
+        # Uses the GIN trigram index on LOWER(canonical_name) for case-insensitive
+        # similarity lookup. Previous version also had LIKE '%...' substring fallbacks,
+        # but those forced full sequential scans of the entities table and caused
+        # TimeoutErrors on banks with 10k+ entities. Lowering the similarity threshold
+        # to 0.15 (from default 0.3) catches most substring relationships while
+        # staying fully index-based.
+        await conn.execute("SET pg_trgm.similarity_threshold = 0.15")
         rows = await conn.fetch(
             f"""
             SELECT DISTINCT ON (e.id)
@@ -299,16 +332,13 @@ class EntityResolver:
             FROM unnest($2::text[]) AS q(query_text)
             JOIN {fq_table("entities")} e ON (
                 e.bank_id = $1
-                AND (
-                    e.canonical_name % q.query_text
-                    OR LOWER(e.canonical_name) LIKE '%' || LOWER(q.query_text) || '%'
-                    OR LOWER(q.query_text) LIKE '%' || LOWER(e.canonical_name) || '%'
-                )
+                AND LOWER(e.canonical_name) % LOWER(q.query_text)
             )
             """,
             bank_id,
             entity_texts,
         )
+        await conn.execute("RESET pg_trgm.similarity_threshold")
 
         # Group candidates by query_text
         all_candidates: dict[str, list] = {t: [] for t in entity_texts}
@@ -477,19 +507,42 @@ class EntityResolver:
             id_by_name: dict[str, str] = {row["name_lower"]: row["id"] for row in inserted_rows}
 
             # Fallback SELECT for names that conflicted (another worker won the race).
-            missing = [n for n, _ in sorted_groups if n not in id_by_name]
-            if missing:
+            #
+            # IMPORTANT: we must let PostgreSQL do the lowercasing on BOTH sides of the
+            # comparison.  Python's str.lower() and PostgreSQL's LOWER() differ for some
+            # Unicode characters — most notably Turkish İ (U+0130):
+            #   Python:     'İstanbul'.lower()  == 'i\u0307stanbul'  (i + combining dot, 2 chars)
+            #   PostgreSQL: LOWER('İstanbul')   == 'istanbul'         (plain i, 1 char)
+            # Passing a Python-lowercased name to "LOWER(canonical_name) = ANY($2::text[])"
+            # would fail to match the stored entity, leaving entity_id as None and causing
+            # a NOT NULL constraint violation on unit_entities.entity_id.
+            #
+            # Fix: pass the original (mixed-case) input names and use
+            # "LOWER(canonical_name) = ANY(SELECT LOWER(n) FROM unnest($2) AS n)" so
+            # PostgreSQL lowercases both sides identically.  The query also returns the
+            # original input_name so we can index id_by_name by Python's lower() of that
+            # name, which is what the assignment loop below uses as its lookup key.
+            missing_original = [g.name for name_lower, g in sorted_groups if name_lower not in id_by_name]
+            if missing_original:
                 existing_rows = await conn.fetch(
                     f"""
-                    SELECT id, LOWER(canonical_name) AS name_lower
-                    FROM {fq_table("entities")}
-                    WHERE bank_id = $1 AND LOWER(canonical_name) = ANY($2::text[])
+                    SELECT e.id, LOWER(e.canonical_name) AS name_lower, inputs.input_name
+                    FROM {fq_table("entities")} e
+                    JOIN (
+                        SELECT LOWER(n) AS input_name_lower, n AS input_name
+                        FROM unnest($2::text[]) AS n
+                    ) AS inputs ON LOWER(e.canonical_name) = inputs.input_name_lower
+                    WHERE e.bank_id = $1
                     """,
                     bank_id,
-                    missing,
+                    missing_original,
                 )
                 for row in existing_rows:
                     id_by_name[row["name_lower"]] = row["id"]
+                    # Also index by Python's lower() of the original input name so the
+                    # assignment loop (which uses Python-lowercased keys) finds it even
+                    # when Python and PostgreSQL produce different lowercase strings.
+                    id_by_name[row["input_name"].lower()] = row["id"]
 
             # Assign entity IDs back and queue one stat per original mention so that
             # flush_pending_stats() increments mention_count by the true mention count,
@@ -757,14 +810,19 @@ class EntityResolver:
             return await self._link_units_to_entities_batch_impl(conn, unit_entity_pairs)
 
     async def _link_units_to_entities_batch_impl(self, conn, unit_entity_pairs: list[tuple[str, str]]):
-        # Batch insert all unit-entity links
-        await conn.executemany(
+        # Sorted bulk insert to prevent deadlocks from inconsistent lock ordering
+        # across concurrent transactions on the unit_entities unique index.
+        sorted_pairs = sorted(unit_entity_pairs)
+        unit_ids = [p[0] for p in sorted_pairs]
+        entity_ids = [p[1] for p in sorted_pairs]
+        await conn.execute(
             f"""
             INSERT INTO {fq_table("unit_entities")} (unit_id, entity_id)
-            VALUES ($1, $2)
+            SELECT u, e FROM unnest($1::uuid[], $2::uuid[]) AS t(u, e)
             ON CONFLICT DO NOTHING
             """,
-            unit_entity_pairs,
+            unit_ids,
+            entity_ids,
         )
 
         # Build map of unit -> entities for co-occurrence calculation

hindsight-api-slim/hindsight_api/engine/interface.py

@@ -240,6 +240,7 @@ class MemoryEngineInterface(ABC):
         bank_id: str,
         *,
         fact_type: str | None = None,
+        delete_bank_profile: bool = True,
         request_context: "RequestContext",
     ) -> dict[str, int]:
         """
@@ -248,6 +249,8 @@ class MemoryEngineInterface(ABC):
         Args:
             bank_id: The memory bank ID.
             fact_type: If specified, only delete memories of this type.
+            delete_bank_profile: If True, also delete the bank profile row itself.
+                If False, only delete memories/entities/documents but preserve the bank.
             request_context: Request context for authentication.
 
         Returns:

hindsight-api-slim/hindsight_api/engine/llm_wrapper.py

@@ -122,10 +122,14 @@ _PROVIDERS_WITHOUT_API_KEY = frozenset(
     {
         "ollama",
         "lmstudio",
+        "llamacpp",
         "openai-codex",
         "claude-code",
         "mock",
+        "none",
         "vertexai",
+        "litellm",
+        "bedrock",
     }
 )
 
@@ -143,6 +147,7 @@ def create_llm_provider(
     reasoning_effort: str,
     groq_service_tier: str | None = None,
     openai_service_tier: str | None = None,
+    extra_body: dict[str, Any] | None = None,
     vertexai_project_id: str | None = None,
     vertexai_region: str | None = None,
     vertexai_credentials: Any = None,
@@ -159,6 +164,7 @@ def create_llm_provider(
         reasoning_effort: Reasoning effort level for supported providers.
         groq_service_tier: Groq service tier (for Groq provider) - "on_demand", "flex", or "auto".
         openai_service_tier: OpenAI service tier (for OpenAI provider) - None (default) or "flex" (50% cheaper).
+        extra_body: Extra body params merged into OpenAI-compatible API calls.
         vertexai_project_id: Vertex AI project ID (for VertexAI provider).
         vertexai_region: Vertex AI region (for VertexAI provider).
         vertexai_credentials: Vertex AI credentials object (for VertexAI provider).
@@ -172,7 +178,10 @@ def create_llm_provider(
         ClaudeCodeLLM,
         CodexLLM,
         GeminiLLM,
+        LiteLLMLLM,
+        LlamaCppLLM,
         MockLLM,
+        NoneLLM,
         OpenAICompatibleLLM,
     )
 
@@ -205,6 +214,15 @@ def create_llm_provider(
             reasoning_effort=reasoning_effort,
         )
 
+    elif provider_lower == "none":
+        return NoneLLM(
+            provider=provider,
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            reasoning_effort=reasoning_effort,
+        )
+
     elif provider_lower in ("gemini", "vertexai"):
         return GeminiLLM(
             provider=provider,
@@ -227,7 +245,45 @@ def create_llm_provider(
             reasoning_effort=reasoning_effort,
         )
 
-    elif provider_lower in ("openai", "groq", "ollama", "lmstudio", "minimax"):
+    elif provider_lower == "litellm":
+        return LiteLLMLLM(
+            provider=provider,
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            reasoning_effort=reasoning_effort,
+        )
+
+    elif provider_lower == "bedrock":
+        # Bedrock is a first-class alias backed by LiteLLM with auto-prefixed model names
+        bedrock_model = model if model.startswith("bedrock/") else f"bedrock/{model}"
+        return LiteLLMLLM(
+            provider=provider,
+            api_key=api_key,
+            base_url=base_url,
+            model=bedrock_model,
+            reasoning_effort=reasoning_effort,
+        )
+
+    elif provider_lower == "llamacpp":
+        from ..config import get_config
+
+        config = get_config()
+        return LlamaCppLLM(
+            provider=provider,
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            reasoning_effort=reasoning_effort,
+            model_path=config.llamacpp_model_path,
+            gpu_layers=config.llamacpp_gpu_layers,
+            context_size=config.llamacpp_context_size,
+            chat_format=config.llamacpp_chat_format,
+            no_grammar=config.llamacpp_no_grammar,
+            extra_args=config.llamacpp_extra_args,
+        )
+
+    elif provider_lower in ("openai", "groq", "ollama", "lmstudio", "minimax", "volcano", "openrouter"):
         return OpenAICompatibleLLM(
             provider=provider,
             api_key=api_key,
@@ -236,6 +292,7 @@ def create_llm_provider(
             reasoning_effort=reasoning_effort,
             groq_service_tier=groq_service_tier,
             openai_service_tier=openai_service_tier,
+            extra_body=extra_body,
         )
 
     else:
@@ -259,6 +316,7 @@ class LLMProvider:
         groq_service_tier: str | None = None,
         openai_service_tier: str | None = None,
         gemini_safety_settings: list | None = None,
+        extra_body: dict[str, Any] | None = None,
     ):
         """
         Initialize LLM provider.
@@ -272,6 +330,7 @@ class LLMProvider:
             groq_service_tier: Groq service tier ("on_demand", "flex", "auto") - from config.
             openai_service_tier: OpenAI service tier (None or "flex") - from config.
             gemini_safety_settings: Safety settings for Gemini/VertexAI providers.
+            extra_body: Extra body params merged into OpenAI-compatible API calls.
         """
         self.provider = provider.lower()
         self.api_key = api_key
@@ -283,6 +342,8 @@ class LLMProvider:
         self.openai_service_tier = openai_service_tier
         # Gemini safety settings (instance default; can be overridden per-request via context var)
         self.gemini_safety_settings = gemini_safety_settings
+        # Extra body params for OpenAI-compatible providers (e.g. chat_template_kwargs)
+        self.extra_body = extra_body
 
         # Validate provider
         valid_providers = [
@@ -292,11 +353,17 @@ class LLMProvider:
             "gemini",
             "anthropic",
             "lmstudio",
+            "llamacpp",
             "vertexai",
             "openai-codex",
             "claude-code",
             "mock",
+            "none",
             "minimax",
+            "litellm",
+            "bedrock",
+            "volcano",
+            "openrouter",
         ]
         if self.provider not in valid_providers:
             raise ValueError(f"Invalid LLM provider: {self.provider}. Must be one of: {', '.join(valid_providers)}")
@@ -311,6 +378,8 @@ class LLMProvider:
                 self.base_url = "http://localhost:1234/v1"
             elif self.provider == "minimax":
                 self.base_url = "https://api.minimax.io/v1"
+            elif self.provider == "openrouter":
+                self.base_url = "https://openrouter.ai/api/v1"
 
         # Prepare Vertex AI config (if applicable)
         vertexai_project_id = None
@@ -375,6 +444,7 @@ class LLMProvider:
             reasoning_effort=self.reasoning_effort,
             groq_service_tier=self.groq_service_tier,
             openai_service_tier=self.openai_service_tier,
+            extra_body=self.extra_body,
             vertexai_project_id=vertexai_project_id,
             vertexai_region=vertexai_region,
             vertexai_credentials=vertexai_credentials,
@@ -466,6 +536,15 @@ class LLMProvider:
             OutputTooLongError: If output exceeds token limits.
             Exception: Re-raises API errors after retries exhausted.
         """
+        # Stage breadcrumb so the worker log shows which LLM call a task is
+        # currently inside; the stage_age field then reveals long JSON-schema
+        # retry loops (e.g. a small model that can't satisfy strict_schema).
+        # No-op outside a worker context.
+        from ..worker.stage import set_stage
+
+        structured = "+structured" if response_format is not None else ""
+        set_stage(f"llm.{self.provider}.{scope}{structured}")
+
         async with _global_llm_semaphore:
             # Delegate to provider implementation
             result = await self._provider_impl.call(
@@ -522,6 +601,10 @@ class LLMProvider:
         Returns:
             LLMToolCallResult with content and/or tool_calls.
         """
+        from ..worker.stage import set_stage
+
+        set_stage(f"llm.{self.provider}.{scope}+tools")
+
         async with _global_llm_semaphore:
             # Delegate to provider implementation
             result = await self._provider_impl.call_with_tools(
@@ -633,7 +716,7 @@ class LLMProvider:
             # Reduce Claude Agent SDK logging verbosity
             import logging as sdk_logging
 
-            from claude_agent_sdk import query  # noqa: F401
+            from claude_agent_sdk import query  # noqa: F401  # type: ignore[unresolved-import]
 
             sdk_logging.getLogger("claude_agent_sdk").setLevel(sdk_logging.WARNING)
             sdk_logging.getLogger("claude_agent_sdk._internal").setLevel(sdk_logging.WARNING)
@@ -665,64 +748,45 @@ class LLMProvider:
         return ConfiguredLLMProvider(self, config.llm_gemini_safety_settings)
 
     async def cleanup(self) -> None:
-        """Clean up resources."""
-        pass
+        """Clean up resources (e.g. stop llamacpp subprocess)."""
+        if self._provider_impl:
+            await self._provider_impl.cleanup()
 
     @classmethod
-    def for_memory(cls) -> "LLMProvider":
-        """Create provider for memory operations from environment variables."""
-        provider = os.getenv("HINDSIGHT_API_LLM_PROVIDER", "groq")
-        api_key = os.getenv("HINDSIGHT_API_LLM_API_KEY", "")
+    def from_env(cls) -> "LLMProvider":
+        """Create provider from environment variables using config.py constants."""
+        from ..config import (
+            DEFAULT_LLM_MODEL,
+            DEFAULT_LLM_PROVIDER,
+            ENV_LLM_API_KEY,
+            ENV_LLM_BASE_URL,
+            ENV_LLM_EXTRA_BODY,
+            ENV_LLM_MODEL,
+            ENV_LLM_PROVIDER,
+        )
 
-        # API key not needed for openai-codex (uses OAuth), claude-code (uses Keychain OAuth),
-        # ollama (local), or vertexai (uses GCP service account credentials)
-        if not api_key and provider not in ("openai-codex", "claude-code", "ollama", "vertexai"):
+        provider = os.getenv(ENV_LLM_PROVIDER, DEFAULT_LLM_PROVIDER)
+        api_key = os.getenv(ENV_LLM_API_KEY, "")
+
+        if not api_key and not requires_api_key(provider):
+            pass  # Provider handles its own auth
+        elif not api_key:
             raise ValueError(
-                "HINDSIGHT_API_LLM_API_KEY environment variable is required (unless using openai-codex or claude-code)"
+                f"{ENV_LLM_API_KEY} environment variable is required (unless using openai-codex, claude-code, or litellm)"
             )
 
-        base_url = os.getenv("HINDSIGHT_API_LLM_BASE_URL", "")
-        model = os.getenv("HINDSIGHT_API_LLM_MODEL", "openai/gpt-oss-120b")
+        base_url = os.getenv(ENV_LLM_BASE_URL, "")
+        model = os.getenv(ENV_LLM_MODEL, DEFAULT_LLM_MODEL)
+        extra_body = json.loads(os.getenv(ENV_LLM_EXTRA_BODY, "null"))
 
-        return cls(provider=provider, api_key=api_key, base_url=base_url, model=model, reasoning_effort="low")
-
-    @classmethod
-    def for_answer_generation(cls) -> "LLMProvider":
-        """Create provider for answer generation. Falls back to memory config if not set."""
-        provider = os.getenv("HINDSIGHT_API_ANSWER_LLM_PROVIDER", os.getenv("HINDSIGHT_API_LLM_PROVIDER", "groq"))
-        api_key = os.getenv("HINDSIGHT_API_ANSWER_LLM_API_KEY", os.getenv("HINDSIGHT_API_LLM_API_KEY", ""))
-
-        # API key not needed for openai-codex (uses OAuth), claude-code (uses Keychain OAuth),
-        # ollama (local), or vertexai (uses GCP service account credentials)
-        if not api_key and provider not in ("openai-codex", "claude-code", "ollama", "vertexai"):
-            raise ValueError(
-                "HINDSIGHT_API_LLM_API_KEY or HINDSIGHT_API_ANSWER_LLM_API_KEY environment variable is required "
-                "(unless using openai-codex or claude-code)"
-            )
-
-        base_url = os.getenv("HINDSIGHT_API_ANSWER_LLM_BASE_URL", os.getenv("HINDSIGHT_API_LLM_BASE_URL", ""))
-        model = os.getenv("HINDSIGHT_API_ANSWER_LLM_MODEL", os.getenv("HINDSIGHT_API_LLM_MODEL", "openai/gpt-oss-120b"))
-
-        return cls(provider=provider, api_key=api_key, base_url=base_url, model=model, reasoning_effort="high")
-
-    @classmethod
-    def for_judge(cls) -> "LLMProvider":
-        """Create provider for judge/evaluator operations. Falls back to memory config if not set."""
-        provider = os.getenv("HINDSIGHT_API_JUDGE_LLM_PROVIDER", os.getenv("HINDSIGHT_API_LLM_PROVIDER", "groq"))
-        api_key = os.getenv("HINDSIGHT_API_JUDGE_LLM_API_KEY", os.getenv("HINDSIGHT_API_LLM_API_KEY", ""))
-
-        # API key not needed for openai-codex (uses OAuth), claude-code (uses Keychain OAuth),
-        # ollama (local), or vertexai (uses GCP service account credentials)
-        if not api_key and provider not in ("openai-codex", "claude-code", "ollama", "vertexai"):
-            raise ValueError(
-                "HINDSIGHT_API_LLM_API_KEY or HINDSIGHT_API_JUDGE_LLM_API_KEY environment variable is required "
-                "(unless using openai-codex or claude-code)"
-            )
-
-        base_url = os.getenv("HINDSIGHT_API_JUDGE_LLM_BASE_URL", os.getenv("HINDSIGHT_API_LLM_BASE_URL", ""))
-        model = os.getenv("HINDSIGHT_API_JUDGE_LLM_MODEL", os.getenv("HINDSIGHT_API_LLM_MODEL", "openai/gpt-oss-120b"))
-
-        return cls(provider=provider, api_key=api_key, base_url=base_url, model=model, reasoning_effort="high")
+        return cls(
+            provider=provider,
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            reasoning_effort="low",
+            extra_body=extra_body,
+        )
 
 
 class ConfiguredLLMProvider:

hindsight-api-slim/hindsight_api/engine/providers/__init__.py

@@ -8,7 +8,20 @@ from .anthropic_llm import AnthropicLLM
 from .claude_code_llm import ClaudeCodeLLM
 from .codex_llm import CodexLLM
 from .gemini_llm import GeminiLLM
+from .litellm_llm import LiteLLMLLM
+from .llamacpp_llm import LlamaCppLLM
 from .mock_llm import MockLLM
+from .none_llm import NoneLLM
 from .openai_compatible_llm import OpenAICompatibleLLM
 
-__all__ = ["AnthropicLLM", "ClaudeCodeLLM", "CodexLLM", "GeminiLLM", "MockLLM", "OpenAICompatibleLLM"]
+__all__ = [
+    "AnthropicLLM",
+    "ClaudeCodeLLM",
+    "CodexLLM",
+    "GeminiLLM",
+    "LlamaCppLLM",
+    "LiteLLMLLM",
+    "MockLLM",
+    "NoneLLM",
+    "OpenAICompatibleLLM",
+]

hindsight-api-slim/hindsight_api/engine/providers/claude_code_llm.py

@@ -68,7 +68,7 @@ class ClaudeCodeLLM(LLMInterface):
             # Reduce Claude Agent SDK logging verbosity
             import logging as sdk_logging
 
-            from claude_agent_sdk import query  # noqa: F401
+            from claude_agent_sdk import query  # noqa: F401  # type: ignore[unresolved-import]
 
             sdk_logging.getLogger("claude_agent_sdk").setLevel(sdk_logging.WARNING)
             sdk_logging.getLogger("claude_agent_sdk._internal").setLevel(sdk_logging.WARNING)
@@ -141,7 +141,12 @@ class ClaudeCodeLLM(LLMInterface):
             OutputTooLongError: If output exceeds token limits (not supported by Claude Agent SDK).
             Exception: Re-raises API errors after retries exhausted.
         """
-        from claude_agent_sdk import AssistantMessage, ClaudeAgentOptions, TextBlock, query
+        from claude_agent_sdk import (  # type: ignore[unresolved-import]
+            AssistantMessage,
+            ClaudeAgentOptions,
+            TextBlock,
+            query,
+        )
 
         start_time = time.time()
 
@@ -326,12 +331,16 @@ class ClaudeCodeLLM(LLMInterface):
             max_retries: Maximum retry attempts.
             initial_backoff: Initial backoff time in seconds.
             max_backoff: Maximum backoff time in seconds.
-            tool_choice: How to choose tools (not used by Claude Agent SDK).
+            tool_choice: How to choose tools - "auto", "none", "required", or specific function dict.
+                - "auto": Model decides whether to call tools (default)
+                - "required": Model must call at least one tool
+                - "none": Model must not call any tools
+                - {"type": "function", "function": {"name": "..."}}: Force specific tool call
 
         Returns:
             LLMToolCallResult with content and/or tool_calls.
         """
-        from claude_agent_sdk import (
+        from claude_agent_sdk import (  # type: ignore[unresolved-import]
             AssistantMessage,
             ClaudeAgentOptions,
             ClaudeSDKClient,
@@ -405,16 +414,57 @@ class ClaudeCodeLLM(LLMInterface):
                 tool_call_id = msg.get("tool_call_id", "")
                 user_content += f"\n\n[Tool result for {tool_call_id}: {content}]"
 
+        # Handle tool_choice parameter to filter tools and adjust instructions
+        # The Claude Agent SDK doesn't have a native tool_choice parameter, so we
+        # enforce it via allowed_tools filtering and system prompt instructions.
+
         # Format tool names for SDK MCP servers: mcp__{server_name}__{tool_name}
         # This is required by the Claude Agent SDK for MCP server tools
         allowed_tool_names = [f"mcp__hindsight_tools__{name}" for name in tool_names]
+        mcp_servers_config = {"hindsight_tools": mcp_server} if sdk_tools else {}
+
+        # Process tool_choice
+        if isinstance(tool_choice, dict) and tool_choice.get("type") == "function":
+            # Force a specific tool: filter allowed_tools to only that tool and add instruction
+            forced_name = tool_choice.get("function", {}).get("name")
+            if forced_name:
+                # Filter to only the forced tool (with MCP prefix)
+                forced_tool_mcp_name = f"mcp__hindsight_tools__{forced_name}"
+                if forced_tool_mcp_name in allowed_tool_names:
+                    allowed_tool_names = [forced_tool_mcp_name]
+                    # Add strong instruction to system prompt
+                    force_instruction = (
+                        f"\n\nIMPORTANT: You MUST call the '{forced_name}' tool. Do not respond with text only."
+                    )
+                    system_prompt += force_instruction
+                    logger.debug(f"Claude Code: Forcing tool call to '{forced_name}'")
+                else:
+                    logger.warning(f"Claude Code: Forced tool '{forced_name}' not found in available tools")
+        elif tool_choice == "required":
+            # Must call at least one tool
+            tool_instruction = (
+                "\n\nIMPORTANT: You MUST call at least one of the available tools. Do not respond with text only."
+            )
+            system_prompt += tool_instruction
+            logger.debug("Claude Code: Tool call required")
+        elif tool_choice == "none":
+            # No tools should be called - disable all tools
+            allowed_tool_names = []
+            mcp_servers_config = {}
+            logger.debug("Claude Code: Tools disabled (tool_choice=none)")
+        # else: tool_choice == "auto" or unspecified - use default behavior (no changes needed)
 
         # Configure SDK options with MCP server
+        # tools=[] disables built-in CLI tools (Read, Write, Bash, ToolSearch, etc.)
+        # Without this, Claude Code CLI defers MCP tools when too many built-in tools
+        # are loaded, forcing Claude to use ToolSearch first — which wastes the max_turns
+        # budget and prevents direct MCP tool calls.
         options = ClaudeAgentOptions(
             system_prompt=system_prompt if system_prompt else None,
-            max_turns=1,  # Single-turn for API-style interactions
-            mcp_servers={"hindsight_tools": mcp_server} if sdk_tools else {},
-            allowed_tools=allowed_tool_names if allowed_tool_names else [],
+            tools=[],  # Disable built-in tools so MCP tools load eagerly
+            max_turns=2,  # Allow tool call + tool result round-trip
+            mcp_servers=mcp_servers_config,
+            allowed_tools=allowed_tool_names,
         )
 
         # Call Claude Agent SDK with retry logic

hindsight-api-slim/hindsight_api/engine/providers/codex_llm.py

@@ -126,6 +126,32 @@ class CodexLLM(LLMInterface):
         }
         return mapping.get(effort.lower(), "auto")
 
+    def _normalize_tool_choice(self, tool_choice: str | dict[str, Any]) -> str | dict[str, Any]:
+        """Normalize forced function tool choice for the Codex Responses API.
+
+        Older agent paths may still pass OpenAI chat-completions style named
+        tool choice payloads such as:
+
+            {"type": "function", "function": {"name": "recall"}}
+
+        Codex Responses expects the named function at the top level instead:
+
+            {"type": "function", "name": "recall"}
+        """
+        if not isinstance(tool_choice, dict):
+            return tool_choice
+        if str(tool_choice.get("type") or "").strip() != "function":
+            return tool_choice
+        function_payload = tool_choice.get("function")
+        if isinstance(function_payload, dict):
+            function_name = str(function_payload.get("name") or "").strip()
+            if function_name:
+                return {"type": "function", "name": function_name}
+        function_name = str(tool_choice.get("name") or "").strip()
+        if function_name:
+            return {"type": "function", "name": function_name}
+        return tool_choice
+
     async def verify_connection(self) -> None:
         """Verify Codex connection by making a simple test call."""
         try:
@@ -140,6 +166,10 @@ class CodexLLM(LLMInterface):
             )
             logger.info(f"Codex LLM verified: {self.model}")
         except Exception as e:
+            # 429 means quota exhausted, not a configuration error — warn but allow startup
+            if "429" in str(e) or "usage_limit_reached" in str(e):
+                logger.warning(f"Codex LLM quota exhausted for {self.model}, continuing startup: {e}")
+                return
             raise RuntimeError(f"Codex LLM connection verification failed for {self.model}: {e}") from e
 
     async def call(
@@ -263,24 +293,27 @@ class CodexLLM(LLMInterface):
                 )
 
                 # Record trace span
-                from hindsight_api.tracing import get_span_recorder
+                try:
+                    from hindsight_api.tracing import get_span_recorder
 
-                # Estimate tokens for tracing
-                estimated_input = sum(len(m.get("content", "")) for m in messages) // 4
-                estimated_output = len(content) // 4
-                span_recorder = get_span_recorder()
-                span_recorder.record_llm_call(
-                    provider=self.provider,
-                    model=self.model,
-                    scope=scope,
-                    messages=messages,
-                    response_content=result if isinstance(result, str) else json.dumps(result),
-                    input_tokens=estimated_input,
-                    output_tokens=estimated_output,
-                    duration=duration,
-                    finish_reason=None,
-                    error=None,
-                )
+                    # Estimate tokens for tracing
+                    estimated_input = sum(len(m.get("content", "")) for m in messages) // 4
+                    estimated_output = len(content) // 4
+                    span_recorder = get_span_recorder()
+                    span_recorder.record_llm_call(
+                        provider=self.provider,
+                        model=self.model,
+                        scope=scope,
+                        messages=messages,
+                        response_content=result if isinstance(result, str) else result.model_dump_json(),
+                        input_tokens=estimated_input,
+                        output_tokens=estimated_output,
+                        duration=duration,
+                        finish_reason=None,
+                        error=None,
+                    )
+                except Exception:
+                    pass  # logging failure must never affect the operation
 
                 if return_usage:
                     # Codex doesn't provide token counts, estimate based on content
@@ -422,7 +455,7 @@ class CodexLLM(LLMInterface):
             max_retries: Maximum retry attempts.
             initial_backoff: Initial backoff time in seconds.
             max_backoff: Maximum backoff time in seconds.
-            tool_choice: How to choose tools - "auto", "none", "required", or specific function.
+            tool_choice: How to choose tools - "auto", "none", "required", or a specific function.
 
         Returns:
             LLMToolCallResult with content and/or tool_calls.
@@ -479,7 +512,7 @@ class CodexLLM(LLMInterface):
             "instructions": system_instruction,
             "input": user_messages,
             "tools": codex_tools,
-            "tool_choice": tool_choice,
+            "tool_choice": self._normalize_tool_choice(tool_choice),
             "parallel_tool_calls": True,
             "reasoning": {"summary": reasoning_summary},
             "store": False,
@@ -526,26 +559,31 @@ class CodexLLM(LLMInterface):
             )
 
             # Record OpenTelemetry span
-            from hindsight_api.tracing import get_span_recorder
+            try:
+                from hindsight_api.tracing import get_span_recorder
 
-            span_recorder = get_span_recorder()
-            # Convert LLMToolCall objects to dicts for span recording
-            tool_calls_dict = (
-                [{"id": tc.id, "name": tc.name, "arguments": tc.arguments} for tc in tool_calls] if tool_calls else None
-            )
-            span_recorder.record_llm_call(
-                provider=self.provider,
-                model=self.model,
-                scope=scope,
-                messages=messages,
-                response_content=content,
-                input_tokens=0,  # Codex doesn't provide token counts
-                output_tokens=0,
-                duration=duration,
-                finish_reason="tool_calls" if tool_calls else "stop",
-                error=None,
-                tool_calls=tool_calls_dict,
-            )
+                span_recorder = get_span_recorder()
+                # Convert LLMToolCall objects to dicts for span recording
+                tool_calls_dict = (
+                    [{"id": tc.id, "name": tc.name, "arguments": tc.arguments} for tc in tool_calls]
+                    if tool_calls
+                    else None
+                )
+                span_recorder.record_llm_call(
+                    provider=self.provider,
+                    model=self.model,
+                    scope=scope,
+                    messages=messages,
+                    response_content=content,
+                    input_tokens=0,  # Codex doesn't provide token counts
+                    output_tokens=0,
+                    duration=duration,
+                    finish_reason="tool_calls" if tool_calls else "stop",
+                    error=None,
+                    tool_calls=tool_calls_dict,
+                )
+            except Exception:
+                pass  # logging failure must never affect the operation
 
             return LLMToolCallResult(
                 content=content,

hindsight-api-slim/hindsight_api/engine/providers/gemini_llm.py

@@ -7,6 +7,7 @@ This provider supports both:
 """
 
 import asyncio
+import base64
 import json
 import logging
 import os
@@ -22,6 +23,7 @@ from hindsight_api.engine.llm_interface import LLMInterface, OutputTooLongError
 from hindsight_api.engine.llm_wrapper import parse_llm_json
 from hindsight_api.engine.response_models import LLMToolCall, LLMToolCallResult, TokenUsage
 from hindsight_api.metrics import get_metrics_collector
+from hindsight_api.worker.stage import set_stage
 
 logger = logging.getLogger(__name__)
 
@@ -241,6 +243,8 @@ class GeminiLLM(LLMInterface):
         last_exception = None
 
         for attempt in range(max_retries + 1):
+            if attempt > 0:
+                set_stage(f"llm.gemini.{scope}.attempt={attempt + 1}/{max_retries + 1}")
             try:
                 response = await asyncio.wait_for(
                     self._client.aio.models.generate_content(
@@ -470,9 +474,12 @@ class GeminiLLM(LLMInterface):
                         fn_name = fn.get("name", "")
                         fn_args_str = fn.get("arguments", "{}")
                         fn_args = parse_llm_json(fn_args_str)
-                        parts.append(
-                            genai_types.Part(function_call=genai_types.FunctionCall(name=fn_name, args=fn_args))
-                        )
+                        thought_signature = tc.get("thought_signature")
+                        fc_kwargs: dict[str, Any] = {"name": fn_name, "args": fn_args}
+                        part_kwargs: dict[str, Any] = {"function_call": genai_types.FunctionCall(**fc_kwargs)}
+                        if thought_signature:
+                            part_kwargs["thought_signature"] = base64.b64decode(thought_signature)
+                        parts.append(genai_types.Part(**part_kwargs))
                     gemini_contents.append(genai_types.Content(role="model", parts=parts))
                 else:
                     gemini_contents.append(genai_types.Content(role="model", parts=[genai_types.Part(text=content)]))
@@ -523,6 +530,8 @@ class GeminiLLM(LLMInterface):
 
         last_exception = None
         for attempt in range(max_retries + 1):
+            if attempt > 0:
+                set_stage(f"llm.gemini.tools.attempt={attempt + 1}/{max_retries + 1}")
             try:
                 response = await asyncio.wait_for(
                     self._client.aio.models.generate_content(
@@ -545,11 +554,16 @@ class GeminiLLM(LLMInterface):
                                 content = part.text
                             if hasattr(part, "function_call") and part.function_call:
                                 fc = part.function_call
+                                _raw_ts = getattr(part, "thought_signature", None)
+                                thought_signature = (
+                                    base64.b64encode(_raw_ts).decode("ascii") if isinstance(_raw_ts, bytes) else _raw_ts
+                                )
                                 tool_calls.append(
                                     LLMToolCall(
                                         id=f"gemini_{len(tool_calls)}",
                                         name=fc.name,
                                         arguments=dict(fc.args) if fc.args else {},
+                                        thought_signature=thought_signature,
                                     )
                                 )
 

hindsight-api-slim/hindsight_api/engine/providers/litellm_llm.py

@@ -0,0 +1,385 @@
+"""
+LiteLLM LLM provider for universal model support.
+
+This provider enables using 100+ LLM providers via the LiteLLM SDK, including:
+- AWS Bedrock (bedrock/anthropic.claude-3-5-sonnet-...)
+- Azure OpenAI (azure/gpt-4o)
+- Together AI (together_ai/meta-llama/...)
+- Any other LiteLLM-supported provider
+
+Uses litellm.acompletion() for async chat completions.
+Authentication for cloud providers (e.g., AWS Bedrock via boto3 credential chain)
+is handled automatically by LiteLLM.
+"""
+
+import asyncio
+import json
+import logging
+import time
+from typing import Any
+
+from hindsight_api.engine.llm_interface import LLMInterface, OutputTooLongError
+from hindsight_api.engine.response_models import LLMToolCall, LLMToolCallResult, TokenUsage
+from hindsight_api.metrics import get_metrics_collector
+from hindsight_api.worker.stage import set_stage
+
+logger = logging.getLogger(__name__)
+
+
+class LiteLLMLLM(LLMInterface):
+    """
+    LLM provider using the LiteLLM SDK for universal model support.
+
+    Supports any model accessible via litellm.acompletion(), including AWS Bedrock,
+    Azure OpenAI, Together AI, Fireworks AI, and more.
+
+    Model names follow LiteLLM conventions with provider prefixes:
+    - bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0
+    - azure/gpt-4o
+    - together_ai/meta-llama/Llama-3-70b-chat-hf
+    - fireworks_ai/accounts/fireworks/models/llama-v3p1-70b-instruct
+    """
+
+    def __init__(
+        self,
+        provider: str,
+        api_key: str,
+        base_url: str,
+        model: str,
+        reasoning_effort: str = "low",
+        timeout: float = 300.0,
+        **kwargs: Any,
+    ):
+        super().__init__(provider, api_key, base_url, model, reasoning_effort, **kwargs)
+        self.timeout = timeout
+        self._litellm: Any = None
+
+        try:
+            import litellm
+
+            self._litellm = litellm
+            # Suppress LiteLLM's verbose logging
+            litellm.suppress_debug_info = True  # type: ignore[assignment]
+            # Drop unsupported params instead of raising errors (e.g. tool_choice on some Bedrock models)
+            litellm.drop_params = True  # type: ignore[assignment]
+            logging.getLogger("LiteLLM").setLevel(logging.WARNING)
+            logger.info(f"LiteLLM SDK initialized for model: {self.model}")
+        except ImportError as e:
+            raise RuntimeError("LiteLLM SDK not installed. Run: uv add litellm or pip install litellm") from e
+
+    async def verify_connection(self) -> None:
+        try:
+            test_messages = [{"role": "user", "content": "test"}]
+            await self.call(
+                messages=test_messages,
+                max_completion_tokens=50,
+                temperature=0.0,
+                scope="verification",
+                max_retries=0,
+            )
+            logger.info("LiteLLM connection verified successfully")
+        except OutputTooLongError:
+            # Truncation is fine for verification — it means the connection works
+            logger.info("LiteLLM connection verified successfully (response truncated)")
+        except Exception as e:
+            logger.error(f"LiteLLM connection verification failed: {e}")
+            raise RuntimeError(f"Failed to verify LiteLLM connection: {e}") from e
+
+    def _build_common_kwargs(
+        self,
+        messages: list[dict[str, Any]],
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+    ) -> dict[str, Any]:
+        """Build common kwargs for litellm calls."""
+        kwargs: dict[str, Any] = {
+            "model": self.model,
+            "messages": messages,
+            "timeout": self.timeout,
+        }
+
+        if self.api_key:
+            kwargs["api_key"] = self.api_key
+        if self.base_url:
+            kwargs["api_base"] = self.base_url
+        if max_completion_tokens is not None:
+            kwargs["max_completion_tokens"] = max_completion_tokens
+        if temperature is not None:
+            kwargs["temperature"] = temperature
+
+        return kwargs
+
+    async def call(
+        self,
+        messages: list[dict[str, str]],
+        response_format: Any | None = None,
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+        scope: str = "memory",
+        max_retries: int = 10,
+        initial_backoff: float = 1.0,
+        max_backoff: float = 60.0,
+        skip_validation: bool = False,
+        strict_schema: bool = False,
+        return_usage: bool = False,
+    ) -> Any:
+        start_time = time.time()
+
+        call_kwargs = self._build_common_kwargs(messages, max_completion_tokens, temperature)
+
+        # Add JSON schema response format if provided
+        if response_format is not None and hasattr(response_format, "model_json_schema"):
+            schema = response_format.model_json_schema()
+            call_kwargs["response_format"] = {
+                "type": "json_schema",
+                "json_schema": {
+                    "name": response_format.__name__ if hasattr(response_format, "__name__") else "response",
+                    "schema": schema,
+                    "strict": strict_schema,
+                },
+            }
+
+        last_exception = None
+
+        for attempt in range(max_retries + 1):
+            if attempt > 0:
+                set_stage(f"llm.litellm.{scope}.attempt={attempt + 1}/{max_retries + 1}")
+            try:
+                response = await self._litellm.acompletion(**call_kwargs)
+
+                content = response.choices[0].message.content or ""
+                finish_reason = response.choices[0].finish_reason
+
+                # Check for length-limited output
+                if finish_reason == "length":
+                    raise OutputTooLongError("LiteLLM response was truncated due to token limit")
+
+                if response_format is not None:
+                    # Strip markdown code fences if present
+                    clean_content = content
+                    if "```json" in content:
+                        clean_content = content.split("```json")[1].split("```")[0].strip()
+                    elif "```" in content:
+                        clean_content = content.split("```")[1].split("```")[0].strip()
+
+                    try:
+                        json_data = json.loads(clean_content)
+                    except json.JSONDecodeError:
+                        json_data = json.loads(content)
+
+                    if skip_validation:
+                        result = json_data
+                    else:
+                        result = response_format.model_validate(json_data)
+                else:
+                    result = content
+
+                # Extract usage
+                input_tokens = getattr(response.usage, "prompt_tokens", 0) or 0
+                output_tokens = getattr(response.usage, "completion_tokens", 0) or 0
+                total_tokens = input_tokens + output_tokens
+
+                # Record metrics
+                duration = time.time() - start_time
+                metrics = get_metrics_collector()
+                metrics.record_llm_call(
+                    provider=self.provider,
+                    model=self.model,
+                    scope=scope,
+                    duration=duration,
+                    input_tokens=input_tokens,
+                    output_tokens=output_tokens,
+                    success=True,
+                )
+
+                # Record trace span
+                from hindsight_api.tracing import _serialize_for_span, get_span_recorder
+
+                span_recorder = get_span_recorder()
+                span_recorder.record_llm_call(
+                    provider=self.provider,
+                    model=self.model,
+                    scope=scope,
+                    messages=messages,
+                    response_content=_serialize_for_span(result),
+                    input_tokens=input_tokens,
+                    output_tokens=output_tokens,
+                    duration=duration,
+                    finish_reason=finish_reason,
+                    error=None,
+                )
+
+                if duration > 10.0:
+                    logger.info(
+                        f"slow llm call: scope={scope}, model={self.provider}/{self.model}, "
+                        f"input_tokens={input_tokens}, output_tokens={output_tokens}, "
+                        f"time={duration:.3f}s"
+                    )
+
+                if return_usage:
+                    token_usage = TokenUsage(
+                        input_tokens=input_tokens,
+                        output_tokens=output_tokens,
+                        total_tokens=total_tokens,
+                    )
+                    return result, token_usage
+                return result
+
+            except OutputTooLongError:
+                raise
+
+            except json.JSONDecodeError as e:
+                last_exception = e
+                if attempt < max_retries:
+                    logger.warning("LiteLLM returned invalid JSON, retrying...")
+                    backoff = min(initial_backoff * (2**attempt), max_backoff)
+                    await asyncio.sleep(backoff)
+                    continue
+                else:
+                    logger.error(f"LiteLLM returned invalid JSON after {max_retries + 1} attempts")
+                    raise
+
+            except Exception as e:
+                error_str = str(e).lower()
+                # Fast fail on auth errors
+                if "401" in error_str or "403" in error_str or "unauthorized" in error_str:
+                    logger.error(f"LiteLLM auth error, not retrying: {e}")
+                    raise
+
+                last_exception = e
+                if attempt < max_retries:
+                    # Retry on rate limits, connection errors, server errors
+                    is_retryable = any(
+                        keyword in error_str
+                        for keyword in ("rate", "limit", "timeout", "connection", "500", "502", "503", "529")
+                    )
+                    if is_retryable:
+                        backoff = min(initial_backoff * (2**attempt), max_backoff)
+                        jitter = backoff * 0.2 * (2 * (time.time() % 1) - 1)
+                        await asyncio.sleep(backoff + jitter)
+                        continue
+
+                logger.error(f"LiteLLM API error after {attempt + 1} attempts: {e}")
+                raise
+
+        if last_exception:
+            raise last_exception
+        raise RuntimeError("LiteLLM call failed after all retries")
+
+    async def call_with_tools(
+        self,
+        messages: list[dict[str, Any]],
+        tools: list[dict[str, Any]],
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+        scope: str = "tools",
+        max_retries: int = 5,
+        initial_backoff: float = 1.0,
+        max_backoff: float = 30.0,
+        tool_choice: str | dict[str, Any] = "auto",
+    ) -> LLMToolCallResult:
+        start_time = time.time()
+
+        call_kwargs = self._build_common_kwargs(messages, max_completion_tokens, temperature)
+        call_kwargs["tools"] = tools
+        call_kwargs["tool_choice"] = tool_choice
+
+        last_exception = None
+        for attempt in range(max_retries + 1):
+            if attempt > 0:
+                set_stage(f"llm.litellm.tools.attempt={attempt + 1}/{max_retries + 1}")
+            try:
+                response = await self._litellm.acompletion(**call_kwargs)
+
+                message = response.choices[0].message
+                content = message.content
+                finish_reason = response.choices[0].finish_reason
+
+                # Extract tool calls
+                tool_calls: list[LLMToolCall] = []
+                if message.tool_calls:
+                    for tc in message.tool_calls:
+                        arguments = tc.function.arguments
+                        if isinstance(arguments, str):
+                            arguments = json.loads(arguments)
+                        tool_calls.append(
+                            LLMToolCall(
+                                id=tc.id,
+                                name=tc.function.name,
+                                arguments=arguments,
+                            )
+                        )
+
+                # Extract usage
+                input_tokens = getattr(response.usage, "prompt_tokens", 0) or 0
+                output_tokens = getattr(response.usage, "completion_tokens", 0) or 0
+
+                # Record metrics
+                duration = time.time() - start_time
+                metrics = get_metrics_collector()
+                metrics.record_llm_call(
+                    provider=self.provider,
+                    model=self.model,
+                    scope=scope,
+                    duration=duration,
+                    input_tokens=input_tokens,
+                    output_tokens=output_tokens,
+                    success=True,
+                )
+
+                # Record trace span
+                from hindsight_api.tracing import get_span_recorder
+
+                span_recorder = get_span_recorder()
+                tool_calls_dict = (
+                    [{"id": tc.id, "name": tc.name, "arguments": tc.arguments} for tc in tool_calls]
+                    if tool_calls
+                    else None
+                )
+                span_recorder.record_llm_call(
+                    provider=self.provider,
+                    model=self.model,
+                    scope=scope,
+                    messages=messages,
+                    response_content=content,
+                    input_tokens=input_tokens,
+                    output_tokens=output_tokens,
+                    duration=duration,
+                    finish_reason=finish_reason,
+                    error=None,
+                    tool_calls=tool_calls_dict,
+                )
+
+                return LLMToolCallResult(
+                    content=content,
+                    tool_calls=tool_calls,
+                    finish_reason=finish_reason or ("tool_calls" if tool_calls else "stop"),
+                    input_tokens=input_tokens,
+                    output_tokens=output_tokens,
+                )
+
+            except Exception as e:
+                error_str = str(e).lower()
+                if "401" in error_str or "403" in error_str or "unauthorized" in error_str:
+                    raise
+
+                last_exception = e
+                if attempt < max_retries:
+                    is_retryable = any(
+                        keyword in error_str
+                        for keyword in ("rate", "limit", "timeout", "connection", "500", "502", "503", "529")
+                    )
+                    if is_retryable:
+                        await asyncio.sleep(min(initial_backoff * (2**attempt), max_backoff))
+                        continue
+
+                logger.error(f"LiteLLM tool call error after {attempt + 1} attempts: {e}")
+                raise
+
+        if last_exception:
+            raise last_exception
+        raise RuntimeError("LiteLLM tool call failed after all retries")
+
+    async def cleanup(self) -> None:
+        """Clean up resources."""
+        pass

hindsight-api-slim/hindsight_api/engine/providers/llamacpp_llm.py

@@ -0,0 +1,428 @@
+"""
+Built-in llama.cpp LLM provider for fully offline operation.
+
+Manages a llama-cpp-python server as a subprocess, downloads GGUF models
+from HuggingFace on first use, and delegates inference to the OpenAI-compatible API.
+
+Usage:
+    HINDSIGHT_API_LLM_PROVIDER=llamacpp
+    HINDSIGHT_API_LLAMACPP_MODEL_PATH=~/.hindsight/models/gemma-4-E2B-it-Q4_K_M.gguf
+    HINDSIGHT_API_LLAMACPP_GPU_LAYERS=-1  # -1 = all layers on GPU
+    HINDSIGHT_API_LLAMACPP_CONTEXT_SIZE=8192
+"""
+
+import asyncio
+import logging
+import os
+import signal
+import socket
+import subprocess
+import sys
+import time
+from pathlib import Path
+from typing import Any
+
+from hindsight_api.engine.llm_interface import LLMInterface
+from hindsight_api.engine.response_models import LLMToolCallResult
+
+logger = logging.getLogger(__name__)
+
+# Default GGUF model for offline mode
+DEFAULT_LLAMACPP_HF_REPO = "bartowski/google_gemma-4-E2B-it-GGUF"
+DEFAULT_LLAMACPP_HF_FILENAME = "google_gemma-4-E2B-it-Q4_K_M.gguf"
+DEFAULT_LLAMACPP_MODEL_ALIAS = "gemma-4-e2b-it"
+
+MODELS_DIR = Path.home() / ".hindsight" / "models"
+
+# Singleton server instance — shared across all LlamaCppLLM instances
+# (retain, reflect, consolidation each create their own LLMProvider,
+# but they should all share one llama.cpp server process)
+_shared_server: "LlamaCppServer | None" = None
+_shared_server_lock = asyncio.Lock()
+
+
+def _find_free_port() -> int:
+    """Find a free TCP port on localhost."""
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.bind(("127.0.0.1", 0))
+        return s.getsockname()[1]
+
+
+def _download_default_model() -> Path:
+    """Download the default GGUF model from HuggingFace if not already cached.
+
+    Returns:
+        Path to the downloaded GGUF file.
+    """
+    try:
+        from huggingface_hub import hf_hub_download
+    except ImportError:
+        raise ImportError(
+            "huggingface-hub is required for automatic model download. "
+            "Install with: pip install 'hindsight-api-slim[local-llm]'"
+        )
+
+    MODELS_DIR.mkdir(parents=True, exist_ok=True)
+    target = MODELS_DIR / DEFAULT_LLAMACPP_HF_FILENAME
+
+    if target.exists():
+        logger.info(f"Using cached model: {target}")
+        return target
+
+    logger.info(
+        f"Downloading {DEFAULT_LLAMACPP_HF_FILENAME} from {DEFAULT_LLAMACPP_HF_REPO} (~3.5 GB, first run only)..."
+    )
+
+    downloaded = hf_hub_download(
+        repo_id=DEFAULT_LLAMACPP_HF_REPO,
+        filename=DEFAULT_LLAMACPP_HF_FILENAME,
+        local_dir=str(MODELS_DIR),
+    )
+
+    logger.info(f"Model downloaded: {downloaded}")
+    return Path(downloaded)
+
+
+def _resolve_model_path(model_path: str | None) -> Path:
+    """Resolve the model path, downloading the default if needed.
+
+    Args:
+        model_path: Explicit path to a GGUF file, or None to use the default.
+
+    Returns:
+        Resolved Path to the GGUF file.
+    """
+    if model_path:
+        p = Path(model_path).expanduser()
+        if not p.exists():
+            raise FileNotFoundError(
+                f"GGUF model not found: {p}\n"
+                f"Set HINDSIGHT_API_LLAMACPP_MODEL_PATH to a valid .gguf file, "
+                f"or remove the setting to auto-download the default model."
+            )
+        return p
+
+    return _download_default_model()
+
+
+class LlamaCppServer:
+    """Manages a llama-cpp-python OpenAI-compatible server as a subprocess."""
+
+    def __init__(
+        self,
+        model_path: Path,
+        port: int,
+        gpu_layers: int = -1,
+        context_size: int = 8192,
+        chat_format: str | None = None,
+        extra_args: str | None = None,
+    ):
+        self.model_path = model_path
+        self.port = port
+        self.gpu_layers = gpu_layers
+        self.context_size = context_size
+        self.chat_format = chat_format
+        self.extra_args = extra_args
+        self._process: subprocess.Popen | None = None
+
+    @property
+    def base_url(self) -> str:
+        return f"http://127.0.0.1:{self.port}/v1"
+
+    async def start(self) -> None:
+        """Start the llama.cpp server subprocess."""
+        cmd = [
+            sys.executable,
+            "-m",
+            "llama_cpp.server",
+            "--model",
+            str(self.model_path),
+            "--host",
+            "127.0.0.1",
+            "--port",
+            str(self.port),
+            "--n_gpu_layers",
+            str(self.gpu_layers),
+            "--n_ctx",
+            str(self.context_size),
+            "--flash_attn",
+            "true",
+            "--n_batch",
+            "2048",
+            # Prompt cache: reuse KV cache for repeated system prompts
+            "--cache",
+            "true",
+        ]
+        # Only pass chat_format if explicitly set (most GGUF models have it embedded)
+        if self.chat_format:
+            cmd.extend(["--chat_format", self.chat_format])
+        # User-provided extra args (e.g. "--type_k 1 --type_v 1 --n_threads 8")
+        if self.extra_args:
+            cmd.extend(self.extra_args.split())
+
+        logger.info(f"Starting llama.cpp server: {' '.join(cmd)}")
+
+        # Write stderr to a log file to avoid pipe buffer deadlock
+        # (llama.cpp outputs a lot of model metadata on stderr during loading)
+        self._log_path = MODELS_DIR / "llamacpp_server.log"
+        self._log_file = open(self._log_path, "w")
+
+        self._process = subprocess.Popen(
+            cmd,
+            stdout=subprocess.DEVNULL,
+            stderr=self._log_file,
+            # Ensure the subprocess is killed when the parent exits
+            preexec_fn=os.setsid if hasattr(os, "setsid") else None,
+        )
+
+        # Wait for the server to be ready
+        await self._wait_for_ready()
+
+    async def _wait_for_ready(self, timeout: float = 120.0) -> None:
+        """Wait for the llama.cpp server to accept connections."""
+        import httpx
+
+        start = time.monotonic()
+        url = f"http://127.0.0.1:{self.port}/v1/models"
+        last_log = start
+
+        while time.monotonic() - start < timeout:
+            # Check if process died
+            if self._process and self._process.poll() is not None:
+                stderr = ""
+                try:
+                    stderr = self._log_path.read_text()[-2000:]
+                except Exception:
+                    pass
+                raise RuntimeError(f"llama.cpp server exited with code {self._process.returncode}.\nstderr: {stderr}")
+
+            try:
+                async with httpx.AsyncClient() as client:
+                    resp = await client.get(url, timeout=5.0)
+                    if resp.status_code == 200:
+                        logger.info(f"llama.cpp server ready on port {self.port}")
+                        return
+            except (httpx.ConnectError, httpx.TimeoutException, httpx.ConnectTimeout):
+                pass
+
+            # Log progress every 15s
+            now = time.monotonic()
+            if now - last_log > 15:
+                elapsed = int(now - start)
+                logger.info(f"Waiting for llama.cpp server to load model... ({elapsed}s)")
+                last_log = now
+
+            await asyncio.sleep(1.0)
+
+        # Timeout — read the log to help debug
+        stderr = ""
+        try:
+            stderr = self._log_path.read_text()[-2000:]
+        except Exception:
+            pass
+        raise TimeoutError(
+            f"llama.cpp server did not become ready within {timeout}s.\n"
+            f"Check model compatibility and available memory.\n"
+            f"Server log: {stderr}"
+        )
+
+    async def stop(self) -> None:
+        """Stop the llama.cpp server subprocess."""
+        if self._process is None:
+            return
+
+        logger.info("Stopping llama.cpp server...")
+        try:
+            # Send SIGTERM to the process group
+            if hasattr(os, "killpg"):
+                os.killpg(os.getpgid(self._process.pid), signal.SIGTERM)
+            else:
+                self._process.terminate()
+
+            # Wait up to 10s for graceful shutdown
+            try:
+                self._process.wait(timeout=10)
+            except subprocess.TimeoutExpired:
+                if hasattr(os, "killpg"):
+                    os.killpg(os.getpgid(self._process.pid), signal.SIGKILL)
+                else:
+                    self._process.kill()
+                self._process.wait(timeout=5)
+        except (ProcessLookupError, OSError):
+            pass  # Process already exited
+        finally:
+            self._process = None
+            if hasattr(self, "_log_file") and self._log_file:
+                self._log_file.close()
+                self._log_file = None
+            logger.info("llama.cpp server stopped")
+
+
+class LlamaCppLLM(LLMInterface):
+    """
+    Built-in llama.cpp provider.
+
+    Manages a llama-cpp-python server subprocess and delegates to OpenAICompatibleLLM
+    for actual inference calls. Handles model downloading and server lifecycle.
+    """
+
+    def __init__(
+        self,
+        provider: str,
+        api_key: str,
+        base_url: str,
+        model: str,
+        reasoning_effort: str = "low",
+        model_path: str | None = None,
+        gpu_layers: int = -1,
+        context_size: int = 8192,
+        chat_format: str | None = None,
+        no_grammar: bool = False,
+        extra_args: str | None = None,
+        **kwargs: Any,
+    ):
+        super().__init__(
+            provider=provider,
+            api_key=api_key or "llamacpp",
+            base_url=base_url or "",
+            model=model or DEFAULT_LLAMACPP_MODEL_ALIAS,
+            reasoning_effort=reasoning_effort,
+        )
+        self._model_path_str = model_path
+        self._gpu_layers = gpu_layers
+        self._context_size = context_size
+        self._chat_format = chat_format
+        self._no_grammar = no_grammar
+        self._extra_args = extra_args
+        self._server: LlamaCppServer | None = None
+        self._delegate: Any = None  # OpenAICompatibleLLM, created after server starts
+        self._initialized = False
+
+    async def _ensure_initialized(self) -> None:
+        """Lazy initialization: download model + start shared server on first use."""
+        if self._initialized:
+            return
+
+        global _shared_server
+
+        from .openai_compatible_llm import OpenAICompatibleLLM
+
+        async with _shared_server_lock:
+            if _shared_server is None:
+                # Resolve and potentially download the model
+                model_path = _resolve_model_path(self._model_path_str)
+                logger.info(f"Using GGUF model: {model_path}")
+
+                # Start the shared llama.cpp server
+                port = _find_free_port()
+                _shared_server = LlamaCppServer(
+                    model_path=model_path,
+                    port=port,
+                    gpu_layers=self._gpu_layers,
+                    context_size=self._context_size,
+                    chat_format=self._chat_format,
+                    extra_args=self._extra_args,
+                )
+                await _shared_server.start()
+
+        self._server = _shared_server
+
+        # Create the delegate that talks to the shared server's OpenAI-compatible API
+        if self._no_grammar:
+            logger.info("Grammar enforcement disabled (HINDSIGHT_API_LLAMACPP_NO_GRAMMAR=true)")
+        self._delegate = OpenAICompatibleLLM(
+            provider="llamacpp",
+            api_key="llamacpp",
+            base_url=self._server.base_url,
+            model=self.model,
+            reasoning_effort=self.reasoning_effort,
+        )
+
+        self._initialized = True
+
+    async def verify_connection(self) -> None:
+        """Verify the llama.cpp server is running and can generate text."""
+        await self._ensure_initialized()
+        # Make a simple test call to verify the model can actually generate
+        await self._delegate.call(
+            messages=[{"role": "user", "content": "Say 'ok'"}],
+            max_completion_tokens=10,
+            max_retries=2,
+            initial_backoff=0.5,
+            max_backoff=2.0,
+            scope="verification",
+        )
+        logger.info("llama.cpp LLM verification passed")
+
+    async def call(
+        self,
+        messages: list[dict[str, str]],
+        response_format: Any | None = None,
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+        scope: str = "memory",
+        max_retries: int = 10,
+        initial_backoff: float = 1.0,
+        max_backoff: float = 60.0,
+        skip_validation: bool = False,
+        strict_schema: bool = False,
+        return_usage: bool = False,
+    ) -> Any:
+        """Delegate call to the OpenAI-compatible API."""
+        await self._ensure_initialized()
+        return await self._delegate.call(
+            messages=messages,
+            response_format=response_format,
+            max_completion_tokens=max_completion_tokens,
+            temperature=temperature,
+            scope=scope,
+            max_retries=max_retries,
+            initial_backoff=initial_backoff,
+            max_backoff=max_backoff,
+            skip_validation=skip_validation,
+            strict_schema=strict_schema,
+            return_usage=return_usage,
+        )
+
+    async def call_with_tools(
+        self,
+        messages: list[dict[str, Any]],
+        tools: list[dict[str, Any]],
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+        scope: str = "tools",
+        max_retries: int = 5,
+        initial_backoff: float = 1.0,
+        max_backoff: float = 30.0,
+        tool_choice: str | dict[str, Any] = "auto",
+    ) -> LLMToolCallResult:
+        """Delegate tool calls to the OpenAI-compatible API."""
+        await self._ensure_initialized()
+        return await self._delegate.call_with_tools(
+            messages=messages,
+            tools=tools,
+            max_completion_tokens=max_completion_tokens,
+            temperature=temperature,
+            scope=scope,
+            max_retries=max_retries,
+            initial_backoff=initial_backoff,
+            max_backoff=max_backoff,
+            tool_choice=tool_choice,
+        )
+
+    async def cleanup(self) -> None:
+        """Stop the shared llama.cpp server."""
+        global _shared_server
+
+        if self._delegate:
+            await self._delegate.cleanup()
+            self._delegate = None
+
+        # Stop the shared server (only the first cleanup call actually stops it)
+        async with _shared_server_lock:
+            if _shared_server is not None:
+                await _shared_server.stop()
+                _shared_server = None
+
+        self._server = None
+        self._initialized = False

hindsight-api-slim/hindsight_api/engine/providers/none_llm.py

@@ -0,0 +1,78 @@
+"""
+No-op LLM provider for chunk-only storage mode.
+
+When the LLM provider is set to "none", the system operates without any LLM dependency.
+Retain uses chunks mode (no fact extraction), and reflect/consolidation are disabled.
+This provider acts as a safety net — if any code path unexpectedly tries to call the LLM,
+it raises a clear error instead of a confusing connection failure.
+"""
+
+import logging
+from typing import Any
+
+from ..llm_interface import LLMInterface
+from ..response_models import LLMToolCallResult
+
+logger = logging.getLogger(__name__)
+
+
+class LLMNotAvailableError(Exception):
+    """Raised when an operation requires an LLM but the provider is set to 'none'."""
+
+    pass
+
+
+class NoneLLM(LLMInterface):
+    """
+    No-op LLM provider that rejects all LLM calls.
+
+    Used when HINDSIGHT_API_LLM_PROVIDER=none to run Hindsight as a chunk store
+    with semantic search but without LLM-based features (fact extraction, reflect,
+    consolidation).
+    """
+
+    async def verify_connection(self) -> None:
+        """No-op — no LLM connection to verify."""
+        logger.debug("NoneLLM: no LLM connection to verify (provider=none)")
+
+    async def call(
+        self,
+        messages: list[dict[str, str]],
+        response_format: Any | None = None,
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+        scope: str = "memory",
+        max_retries: int = 10,
+        initial_backoff: float = 1.0,
+        max_backoff: float = 60.0,
+        skip_validation: bool = False,
+        strict_schema: bool = False,
+        return_usage: bool = False,
+    ) -> Any:
+        """Raise LLMNotAvailableError — no LLM is configured."""
+        raise LLMNotAvailableError(
+            "LLM provider is set to 'none'. This operation requires an LLM. "
+            "Set HINDSIGHT_API_LLM_PROVIDER to a real provider (e.g., openai, anthropic, gemini)."
+        )
+
+    async def call_with_tools(
+        self,
+        messages: list[dict[str, Any]],
+        tools: list[dict[str, Any]],
+        max_completion_tokens: int | None = None,
+        temperature: float | None = None,
+        scope: str = "tools",
+        max_retries: int = 5,
+        initial_backoff: float = 1.0,
+        max_backoff: float = 30.0,
+        tool_choice: str | dict[str, Any] = "auto",
+    ) -> LLMToolCallResult:
+        """Raise LLMNotAvailableError — no LLM is configured."""
+        raise LLMNotAvailableError(
+            "LLM provider is set to 'none'. This operation requires an LLM. "
+            "Set HINDSIGHT_API_LLM_PROVIDER to a real provider (e.g., openai, anthropic, gemini)."
+        )
+
+    async def cleanup(self) -> None:
+        """No-op — nothing to clean up."""
+        pass

hindsight-api-slim/hindsight_api/engine/providers/openai_compatible_llm.py

@@ -6,7 +6,7 @@ This provider handles all OpenAI API-compatible models including:
 - Groq: Fast inference with seed control and service tiers
 - Ollama: Local models with native streaming API support
 - LMStudio: Local models with OpenAI-compatible API
-- MiniMax: MiniMax-M2.5 models with 204K context window
+- MiniMax: MiniMax-M2.7 models with 1M context window
 
 Features:
 - Reasoning models with extended thinking (o1, o3, GPT-5 families)
@@ -24,6 +24,7 @@ import os
 import re
 import time
 from typing import Any
+from urllib.parse import parse_qs, urlparse, urlunparse
 
 import httpx
 from openai import APIConnectionError, APIStatusError, AsyncOpenAI, LengthFinishReasonError
@@ -32,6 +33,7 @@ from hindsight_api.config import DEFAULT_LLM_TIMEOUT, ENV_LLM_TIMEOUT
 from hindsight_api.engine.llm_interface import LLMInterface, OutputTooLongError
 from hindsight_api.engine.response_models import LLMToolCall, LLMToolCallResult, TokenUsage
 from hindsight_api.metrics import get_metrics_collector
+from hindsight_api.worker.stage import set_stage
 
 logger = logging.getLogger(__name__)
 
@@ -39,6 +41,25 @@ logger = logging.getLogger(__name__)
 DEFAULT_LLM_SEED = 4242
 
 
+def _strip_code_fences(content: str) -> str:
+    """Strip markdown code fences from LLM response if present.
+
+    Many LLM providers (MiniMax, some Ollama models, Claude via proxies)
+    wrap JSON responses in ```json ... ``` fences even when json_object
+    response format is requested. This strips the fences while preserving
+    the JSON content inside. Returns the original content unchanged if
+    no fences are detected.
+    """
+    if "```" not in content:
+        return content
+    try:
+        if "```json" in content:
+            return content.split("```json")[1].split("```")[0].strip()
+        return content.split("```")[1].split("```")[0].strip()
+    except (IndexError, ValueError):
+        return content
+
+
 class OpenAICompatibleLLM(LLMInterface):
     """
     LLM provider for OpenAI-compatible APIs.
@@ -48,7 +69,7 @@ class OpenAICompatibleLLM(LLMInterface):
     - Groq: Fast inference with seed control and service tiers
     - Ollama: Local models with native streaming API for better structured output
     - LMStudio: Local models with OpenAI-compatible API
-    - MiniMax: MiniMax-M2.5 models via OpenAI-compatible API (https://api.minimax.io/v1)
+    - MiniMax: MiniMax-M2.7 models via OpenAI-compatible API (https://api.minimax.io/v1)
     """
 
     def __init__(
@@ -60,6 +81,7 @@ class OpenAICompatibleLLM(LLMInterface):
         reasoning_effort: str = "low",
         timeout: float | None = None,
         groq_service_tier: str | None = None,
+        extra_body: dict[str, Any] | None = None,
         **kwargs: Any,
     ):
         """
@@ -73,12 +95,13 @@ class OpenAICompatibleLLM(LLMInterface):
             reasoning_effort: Reasoning effort level for supported models ("low", "medium", "high").
             timeout: Request timeout in seconds (uses env var or 300s default).
             groq_service_tier: Groq service tier ("on_demand", "flex", "auto").
+            extra_body: Extra body params merged into every API call.
             **kwargs: Additional provider-specific parameters.
         """
         super().__init__(provider, api_key, base_url, model, reasoning_effort, **kwargs)
 
         # Validate provider
-        valid_providers = ["openai", "groq", "ollama", "lmstudio", "minimax"]
+        valid_providers = ["openai", "groq", "ollama", "lmstudio", "llamacpp", "minimax", "volcano", "openrouter"]
         if self.provider not in valid_providers:
             raise ValueError(f"OpenAICompatibleLLM only supports: {', '.join(valid_providers)}. Got: {self.provider}")
 
@@ -92,26 +115,38 @@ class OpenAICompatibleLLM(LLMInterface):
                 self.base_url = "http://localhost:1234/v1"
             elif self.provider == "minimax":
                 self.base_url = "https://api.minimax.io/v1"
+            elif self.provider == "openrouter":
+                self.base_url = "https://openrouter.ai/api/v1"
 
         # For ollama/lmstudio, use dummy key if not provided
         if self.provider in ("ollama", "lmstudio") and not self.api_key:
             self.api_key = "local"
 
         # Validate API key for cloud providers
-        if self.provider in ("openai", "groq", "minimax") and not self.api_key:
+        if self.provider in ("openai", "groq", "minimax", "openrouter") and not self.api_key:
             raise ValueError(f"API key is required for {self.provider}")
 
         # Service tier configuration (from config, not env vars)
         self.groq_service_tier = groq_service_tier
         self.openai_service_tier = kwargs.get("openai_service_tier")
+        # User-configured extra body params (merged into every API call)
+        self._config_extra_body = extra_body or {}
 
         # Get timeout config
         self.timeout = timeout or float(os.getenv(ENV_LLM_TIMEOUT, str(DEFAULT_LLM_TIMEOUT)))
 
-        # Create OpenAI client
+        # Create OpenAI client — extract query params from base_url (e.g. Azure api-version)
         client_kwargs: dict[str, Any] = {"api_key": self.api_key, "max_retries": 0}
         if self.base_url:
-            client_kwargs["base_url"] = self.base_url
+            parsed = urlparse(self.base_url)
+            if parsed.query:
+                clean_url = urlunparse(parsed._replace(query=""))
+                client_kwargs["base_url"] = clean_url
+                default_query = {k: v[0] for k, v in parse_qs(parsed.query).items()}
+                client_kwargs["default_query"] = default_query
+                self.base_url = clean_url
+            else:
+                client_kwargs["base_url"] = self.base_url
         if self.timeout:
             client_kwargs["timeout"] = self.timeout
 
@@ -159,6 +194,36 @@ class OpenAICompatibleLLM(LLMInterface):
 
         return None
 
+    def _max_tokens_param_name(self) -> str:
+        """Return the correct parameter name for limiting response tokens.
+
+        Native OpenAI, Azure OpenAI, Groq, and llamacpp accept 'max_completion_tokens'.
+        Mistral and other OpenAI-compatible endpoints that haven't adopted the newer
+        parameter name require 'max_tokens', so when the openai provider is configured
+        with a non-Azure custom base_url we fall back to the widely-supported
+        'max_tokens'.
+
+        Reasoning models (GPT-5, o1, o3) only accept 'max_completion_tokens' and reject
+        'max_tokens' outright, so they always use the new parameter name regardless of
+        base_url.
+        """
+        # Reasoning models (GPT-5, o1, o3, ...) only accept max_completion_tokens.
+        # Azure OpenAI + GPT-5 is the canonical example: issue #978.
+        if self._supports_reasoning_model():
+            return "max_completion_tokens"
+        # Native OpenAI (no custom base URL), Groq, and llamacpp use max_completion_tokens
+        if self.provider in ("groq", "llamacpp"):
+            return "max_completion_tokens"
+        if self.provider == "openai" and not self.base_url:
+            return "max_completion_tokens"
+        # Azure OpenAI is fully OpenAI-API-compatible — detect it by hostname so users
+        # can keep provider=openai + an Azure base_url (the documented setup).
+        if self.provider == "openai" and self.base_url and ".openai.azure.com" in self.base_url:
+            return "max_completion_tokens"
+        # openai with custom base_url, ollama, lmstudio, minimax, volcano —
+        # use the widely-supported max_tokens
+        return "max_tokens"
+
     async def call(
         self,
         messages: list[dict[str, str]],
@@ -231,9 +296,7 @@ class OpenAICompatibleLLM(LLMInterface):
             # For reasoning models, enforce minimum to ensure space for reasoning + output
             if is_reasoning_model and max_completion_tokens < 16000:
                 max_completion_tokens = 16000
-            call_params["max_completion_tokens"] = max_completion_tokens
-
-        # Temperature - reasoning models don't support custom temperature
+            call_params[self._max_tokens_param_name()] = max_completion_tokens
         if temperature is not None and not is_reasoning_model:
             # MiniMax requires temperature in (0.0, 1.0] — clamp accordingly
             if self.provider == "minimax":
@@ -245,17 +308,17 @@ class OpenAICompatibleLLM(LLMInterface):
             call_params["reasoning_effort"] = self.reasoning_effort
 
         # Provider-specific parameters
+        extra_body: dict[str, Any] = {**self._config_extra_body}
         if self.provider == "groq":
             call_params["seed"] = DEFAULT_LLM_SEED
-            extra_body: dict[str, Any] = {}
             # Add service_tier if configured
             if self.groq_service_tier:
                 extra_body["service_tier"] = self.groq_service_tier
             # Add reasoning parameters for reasoning models
             if is_reasoning_model:
                 extra_body["include_reasoning"] = False
-            if extra_body:
-                call_params["extra_body"] = extra_body
+        if extra_body:
+            call_params["extra_body"] = extra_body
 
         # Prepare response format ONCE before retry loop
         if response_format is not None:
@@ -288,13 +351,23 @@ class OpenAICompatibleLLM(LLMInterface):
                         first_msg = call_params["messages"][0]
                         if isinstance(first_msg, dict) and isinstance(first_msg.get("content"), str):
                             first_msg["content"] = schema_msg + "\n\n" + first_msg["content"]
-                if self.provider not in ("lmstudio", "ollama"):
-                    # LM Studio and Ollama don't support json_object response format reliably
+                # Providers that skip json_object grammar enforcement
+                skip_grammar = self.provider in ("lmstudio", "ollama", "volcano")
+                if self.provider == "llamacpp":
+                    from hindsight_api.config import get_config
+
+                    skip_grammar = get_config().llamacpp_no_grammar
+                if not skip_grammar:
                     call_params["response_format"] = {"type": "json_object"}
 
         last_exception = None
 
         for attempt in range(max_retries + 1):
+            # Surface attempt count in worker stage so JSON-schema retry loops
+            # are visible from logs (small models on strict structured output
+            # often loop here). Cheap no-op outside worker context.
+            if attempt > 0:
+                set_stage(f"llm.{self.provider}.{scope}.attempt={attempt + 1}/{max_retries + 1}")
             try:
                 if response_format is not None:
                     response = await self._client.chat.completions.create(**call_params)
@@ -313,20 +386,14 @@ class OpenAICompatibleLLM(LLMInterface):
                         if len(content) < original_len:
                             logger.debug(f"Stripped {original_len - len(content)} chars of reasoning tokens")
 
-                    # For local models, they may wrap JSON in markdown code blocks
-                    if self.provider in ("lmstudio", "ollama"):
-                        clean_content = content
-                        if "```json" in content:
-                            clean_content = content.split("```json")[1].split("```")[0].strip()
-                        elif "```" in content:
-                            clean_content = content.split("```")[1].split("```")[0].strip()
-                        try:
-                            json_data = json.loads(clean_content)
-                        except json.JSONDecodeError:
-                            # Fallback to parsing raw content
-                            json_data = json.loads(content)
-                    else:
-                        # Log raw LLM response for debugging JSON parse issues
+                    # Strip markdown code fences if present — any provider may
+                    # produce these (confirmed with MiniMax, some Ollama models,
+                    # Claude via proxies). No-op when content is already bare JSON.
+                    clean_content = _strip_code_fences(content)
+                    try:
+                        json_data = json.loads(clean_content)
+                    except json.JSONDecodeError:
+                        # Fallback to parsing raw content in case stripping was wrong
                         try:
                             json_data = json.loads(content)
                         except json.JSONDecodeError as json_err:
@@ -551,7 +618,7 @@ class OpenAICompatibleLLM(LLMInterface):
         }
 
         if max_completion_tokens is not None:
-            call_params["max_completion_tokens"] = max_completion_tokens
+            call_params[self._max_tokens_param_name()] = max_completion_tokens
         if temperature is not None:
             # MiniMax requires temperature in (0.0, 1.0] — clamp accordingly
             if self.provider == "minimax":
@@ -559,12 +626,17 @@ class OpenAICompatibleLLM(LLMInterface):
             call_params["temperature"] = temperature
 
         # Provider-specific parameters
+        extra_body: dict[str, Any] = {**self._config_extra_body}
         if self.provider == "groq":
             call_params["seed"] = DEFAULT_LLM_SEED
+        if extra_body:
+            call_params["extra_body"] = extra_body
 
         last_exception = None
 
         for attempt in range(max_retries + 1):
+            if attempt > 0:
+                set_stage(f"llm.{self.provider}.tools.attempt={attempt + 1}/{max_retries + 1}")
             try:
                 response = await self._client.chat.completions.create(**call_params)
 
@@ -714,6 +786,8 @@ class OpenAICompatibleLLM(LLMInterface):
 
         async with httpx.AsyncClient(timeout=300.0) as client:
             for attempt in range(max_retries + 1):
+                if attempt > 0:
+                    set_stage(f"llm.ollama_native.{scope}.attempt={attempt + 1}/{max_retries + 1}")
                 try:
                     response = await client.post(native_url, json=payload)
                     response.raise_for_status()
@@ -721,26 +795,33 @@ class OpenAICompatibleLLM(LLMInterface):
                     result = response.json()
                     content = result.get("message", {}).get("content", "")
 
-                    # Parse JSON response
+                    # Strip markdown code fences if present (safety net —
+                    # Ollama with schema enforcement usually returns bare JSON,
+                    # but some models may still wrap in fences)
+                    clean_content = _strip_code_fences(content)
                     try:
-                        json_data = json.loads(content)
-                    except json.JSONDecodeError as json_err:
-                        content_preview = content[:500] if content else "<empty>"
-                        if content and len(content) > 700:
-                            content_preview = f"{content[:500]}...TRUNCATED...{content[-200:]}"
-                        logger.warning(
-                            f"Ollama JSON parse error (attempt {attempt + 1}/{max_retries + 1}): {json_err}\n"
-                            f"  Model: ollama/{self.model}\n"
-                            f"  Content length: {len(content) if content else 0} chars\n"
-                            f"  Content preview: {content_preview!r}"
-                        )
-                        if attempt < max_retries:
-                            backoff = min(initial_backoff * (2**attempt), max_backoff)
-                            await asyncio.sleep(backoff)
-                            last_exception = json_err
-                            continue
-                        else:
-                            raise
+                        json_data = json.loads(clean_content)
+                    except json.JSONDecodeError:
+                        # Fallback to raw content
+                        try:
+                            json_data = json.loads(content)
+                        except json.JSONDecodeError as json_err:
+                            content_preview = content[:500] if content else "<empty>"
+                            if content and len(content) > 700:
+                                content_preview = f"{content[:500]}...TRUNCATED...{content[-200:]}"
+                            logger.warning(
+                                f"Ollama JSON parse error (attempt {attempt + 1}/{max_retries + 1}): {json_err}\n"
+                                f"  Model: ollama/{self.model}\n"
+                                f"  Content length: {len(content) if content else 0} chars\n"
+                                f"  Content preview: {content_preview!r}"
+                            )
+                            if attempt < max_retries:
+                                backoff = min(initial_backoff * (2**attempt), max_backoff)
+                                await asyncio.sleep(backoff)
+                                last_exception = json_err
+                                continue
+                            else:
+                                raise
 
                     # Extract token usage from Ollama response
                     duration = time.time() - start_time

hindsight-api-slim/hindsight_api/engine/query_analyzer.py

@@ -137,7 +137,21 @@ class DateparserQueryAnalyzer(QueryAnalyzer):
             "RETURN_AS_TIMEZONE_AWARE": False,
         }
 
-        results = self._search_dates(query, settings=settings)
+        # Wrap dateparser in a defensive try/except. dateparser has been
+        # observed to crash with internal errors (e.g., IndexError from
+        # locale.translate_search) on certain query inputs. A parser bug
+        # should not bring down the whole search/consolidation pipeline —
+        # treat any failure as "no temporal constraint found" so the caller
+        # can fall back to non-temporal retrieval.
+        try:
+            results = self._search_dates(query, settings=settings)
+        except Exception as e:
+            logger.warning(
+                "dateparser raised %s on query (treating as no temporal constraint): %s",
+                type(e).__name__,
+                e,
+            )
+            return QueryAnalysis(temporal_constraint=None)
 
         if not results:
             return QueryAnalysis(temporal_constraint=None)

hindsight-api-slim/hindsight_api/engine/reflect/agent.py

@@ -316,6 +316,8 @@ async def run_reflect_agent(
     response_schema: dict | None = None,
     directives: list[dict[str, Any]] | None = None,
     has_mental_models: bool = False,
+    include_observations: bool = True,
+    include_recall: bool = True,
     budget: str | None = None,
     max_context_tokens: int = 100_000,
 ) -> ReflectAgentResult:
@@ -355,7 +357,14 @@ async def run_reflect_agent(
     directive_rules = _extract_directive_rules(directives) if directives else None
 
     # Get tools for this agent (with directive compliance field if directives exist)
-    tools = get_reflect_tools(directive_rules=directive_rules)
+    tools = get_reflect_tools(
+        directive_rules=directive_rules,
+        include_mental_models=has_mental_models,
+        include_observations=include_observations,
+        include_recall=include_recall,
+    )
+    # Build set of enabled tool names to guard against LLM hallucinating disabled tool calls
+    enabled_tools: frozenset[str] = frozenset(t["function"]["name"] for t in tools if t.get("type") == "function")
 
     # Build initial messages (directives are injected into system prompt at START and END)
     system_prompt = build_system_prompt_for_tools(
@@ -538,19 +547,18 @@ async def run_reflect_agent(
         llm_start = time.time()
 
         # Determine tool_choice for this iteration.
-        # Force the full hierarchical retrieval path before allowing auto:
-        # With mental models:
-        #   0 → search_mental_models, 1 → search_observations, 2 → recall, 3+ → auto
-        # Without mental models:
-        #   0 → search_observations, 1 → recall, 2+ → auto
-        if iteration == 0 and has_mental_models:
-            iter_tool_choice: str | dict = {"type": "function", "function": {"name": "search_mental_models"}}
-        elif iteration == 0:
-            iter_tool_choice = {"type": "function", "function": {"name": "search_observations"}}
-        elif iteration == 1 and has_mental_models:
-            iter_tool_choice = {"type": "function", "function": {"name": "search_observations"}}
-        elif iteration == 1 or (iteration == 2 and has_mental_models):
-            iter_tool_choice = {"type": "function", "function": {"name": "recall"}}
+        # Force the full hierarchical retrieval path (only for enabled tools) before allowing auto.
+        # Build the forced sequence from the tools that are actually enabled.
+        forced_sequence = []
+        if has_mental_models:
+            forced_sequence.append("search_mental_models")
+        if include_observations:
+            forced_sequence.append("search_observations")
+        if include_recall:
+            forced_sequence.append("recall")
+
+        if iteration < len(forced_sequence):
+            iter_tool_choice: str | dict = {"type": "function", "function": {"name": forced_sequence[iteration]}}
         else:
             iter_tool_choice = "auto"
 
@@ -769,7 +777,17 @@ async def run_reflect_agent(
         # Execute other tools in parallel (exclude done tool in all its format variants)
         other_tools = [tc for tc in result.tool_calls if not _is_done_tool(tc.name)]
         if other_tools:
-            # Add assistant message with tool calls
+            # Partition into enabled vs hallucinated (not in enabled_tools set)
+            allowed_tools = []
+            hallucinated_tools = []
+            for tc in other_tools:
+                norm = _normalize_tool_name(tc.name)
+                if enabled_tools is not None and norm not in enabled_tools and norm not in ("done", "expand"):
+                    hallucinated_tools.append(tc)
+                else:
+                    allowed_tools.append(tc)
+
+            # Build assistant message with all tool calls (LLM requires them for history)
             messages.append(
                 {
                     "role": "assistant",
@@ -777,6 +795,23 @@ async def run_reflect_agent(
                 }
             )
 
+            # Immediately reject hallucinated tool calls without adding to trace
+            for tc in hallucinated_tools:
+                messages.append(
+                    {
+                        "role": "tool",
+                        "tool_call_id": tc.id,
+                        "name": tc.name,
+                        "content": json.dumps(
+                            {
+                                "error": f"Tool '{_normalize_tool_name(tc.name)}' is not available. Use only the tools provided to you."
+                            }
+                        ),
+                    }
+                )
+
+            other_tools = allowed_tools
+
             # Execute tools in parallel
             tool_tasks = [
                 _execute_tool_with_timing(
@@ -785,6 +820,7 @@ async def run_reflect_agent(
                     search_observations_fn,
                     recall_fn,
                     expand_fn,
+                    enabled_tools=enabled_tools,
                 )
                 for tc in other_tools
             ]
@@ -895,7 +931,7 @@ async def run_reflect_agent(
 
 def _tool_call_to_dict(tc: "LLMToolCall") -> dict[str, Any]:
     """Convert LLMToolCall to OpenAI message format."""
-    return {
+    d: dict[str, Any] = {
         "id": tc.id,
         "type": "function",
         "function": {
@@ -903,6 +939,9 @@ def _tool_call_to_dict(tc: "LLMToolCall") -> dict[str, Any]:
             "arguments": json.dumps(tc.arguments),
         },
     }
+    if tc.thought_signature is not None:
+        d["thought_signature"] = tc.thought_signature
+    return d
 
 
 async def _process_done_tool(
@@ -971,6 +1010,7 @@ async def _execute_tool_with_timing(
     search_observations_fn: Callable[[str, int], Awaitable[dict[str, Any]]],
     recall_fn: Callable[[str, int, int], Awaitable[dict[str, Any]]],
     expand_fn: Callable[[list[str], str], Awaitable[dict[str, Any]]],
+    enabled_tools: frozenset[str] | None = None,
 ) -> tuple[dict[str, Any], int]:
     """Execute a tool call and return result with timing."""
     from hindsight_api.tracing import get_tracer
@@ -1004,6 +1044,7 @@ async def _execute_tool_with_timing(
                 search_observations_fn,
                 recall_fn,
                 expand_fn,
+                enabled_tools=enabled_tools,
             )
 
             # Set success attributes
@@ -1043,11 +1084,16 @@ async def _execute_tool(
     search_observations_fn: Callable[[str, int], Awaitable[dict[str, Any]]],
     recall_fn: Callable[[str, int, int], Awaitable[dict[str, Any]]],
     expand_fn: Callable[[list[str], str], Awaitable[dict[str, Any]]],
+    enabled_tools: frozenset[str] | None = None,
 ) -> dict[str, Any]:
     """Execute a single tool by name."""
     # Normalize tool name for various LLM output formats
     tool_name = _normalize_tool_name(tool_name)
 
+    # Guard against LLMs hallucinating calls to tools that were not provided
+    if enabled_tools is not None and tool_name not in enabled_tools and tool_name not in ("done", "expand"):
+        return {"error": f"Tool '{tool_name}' is not available. Use only the tools provided to you."}
+
     if tool_name == "search_mental_models":
         query = args.get("query")
         if not query:

hindsight-api-slim/hindsight_api/engine/reflect/tools.py

@@ -9,6 +9,7 @@ Implements hierarchical retrieval:
 
 import logging
 import uuid
+from dataclasses import replace
 from datetime import datetime, timezone
 from typing import TYPE_CHECKING, Any
 
@@ -29,6 +30,7 @@ async def tool_search_mental_models(
     max_results: int = 5,
     tags: list[str] | None = None,
     tags_match: str = "any",
+    tag_groups: "list | None" = None,
     exclude_ids: list[str] | None = None,
     pending_consolidation: int = 0,
 ) -> dict[str, Any]:
@@ -52,7 +54,7 @@ async def tool_search_mental_models(
         Dict with matching mental models including content and freshness info
     """
     from ..memory_engine import fq_table
-    from ..search.tags import build_tags_where_clause
+    from ..search.tags import build_tag_groups_where_clause, build_tags_where_clause
 
     # Build filters dynamically
     filters = ""
@@ -65,6 +67,11 @@ async def tool_search_mental_models(
         filters += f" {tag_clause}"
         params.extend(tag_params)
 
+    if tag_groups:
+        groups_clause, groups_params, next_param = build_tag_groups_where_clause(tag_groups, next_param)
+        filters += f" {groups_clause}"
+        params.extend(groups_params)
+
     if exclude_ids:
         filters += f" AND id != ALL(${next_param}::text[])"
         params.append(exclude_ids)
@@ -125,11 +132,13 @@ async def tool_search_observations(
     max_tokens: int = 5000,
     tags: list[str] | None = None,
     tags_match: str = "any",
+    tag_groups: "list | None" = None,
     last_consolidated_at: datetime | None = None,
     pending_consolidation: int = 0,
+    source_facts_max_tokens: int = -1,
 ) -> dict[str, Any]:
     """
-    Search consolidated observations using recall with include_source_facts.
+    Search consolidated observations using recall.
 
     Observations are auto-generated from memories. Returns freshness info
     so the agent knows if it should also verify with recall().
@@ -144,23 +153,35 @@ async def tool_search_observations(
         tags_match: How to match tags - "any" (OR), "all" (AND)
         last_consolidated_at: When consolidation last ran (for staleness check)
         pending_consolidation: Number of memories waiting to be consolidated
+        source_facts_max_tokens: Token budget for source facts (-1 = disabled, 0+ = enabled with limit)
 
     Returns:
         Dict with matching observations including freshness info and source memories
     """
+    include_source_facts = source_facts_max_tokens != -1
+    recall_kwargs: dict[str, Any] = {}
+    if include_source_facts and source_facts_max_tokens > 0:
+        recall_kwargs["max_source_facts_tokens"] = source_facts_max_tokens
+
+    # Use an internal request context so this recall is not billed as a
+    # user-facing operation. The reflect caller is already billed for the
+    # overall reflect operation; double-billing the sub-recalls would
+    # overcharge the customer.
+    internal_ctx = replace(request_context, internal=True)
     result = await memory_engine.recall_async(
         bank_id=bank_id,
         query=query,
         fact_type=["observation"],
         max_tokens=max_tokens,
         enable_trace=False,
-        request_context=request_context,
+        request_context=internal_ctx,
         tags=tags,
         tags_match=tags_match,
-        include_source_facts=True,
-        max_source_facts_tokens=-1,  # No token limit — include all source facts
+        tag_groups=tag_groups,
+        include_source_facts=include_source_facts,
         _connection_budget=1,
         _quiet=True,
+        **recall_kwargs,
     )
 
     is_stale = pending_consolidation > 0
@@ -189,8 +210,11 @@ async def tool_recall(
     max_tokens: int = 2048,
     tags: list[str] | None = None,
     tags_match: str = "any",
+    tag_groups: "list | None" = None,
     connection_budget: int = 1,
     max_chunk_tokens: int = 1000,
+    fact_types: list[str] | None = None,
+    include_chunks: bool = True,
 ) -> dict[str, Any]:
     """
     Search memories using TEMPR retrieval.
@@ -207,21 +231,26 @@ async def tool_recall(
         tags: Filter by tags (includes untagged memories)
         tags_match: How to match tags - "any" (OR), "all" (AND), or "exact"
         connection_budget: Max DB connections for this recall (default 1 for internal ops)
-        max_chunk_tokens: Maximum tokens for raw source chunk text (default 1000, always included)
+        max_chunk_tokens: Maximum tokens for raw source chunk text (default 1000)
+        fact_types: Optional filter for fact types to retrieve. Defaults to ["experience", "world"].
+        include_chunks: Whether to fetch raw chunk text alongside facts (default True).
 
     Returns:
-        Dict with list of matching memories including raw chunk text
+        Dict with list of matching memories including raw chunk text (when include_chunks)
     """
-    include_chunks = True
+    # Only world/experience are valid for raw recall (observation is handled by search_observations)
+    recall_fact_type = [ft for ft in (fact_types or ["experience", "world"]) if ft in ("world", "experience")]
+    internal_ctx = replace(request_context, internal=True)
     result = await memory_engine.recall_async(
         bank_id=bank_id,
         query=query,
-        fact_type=["experience", "world"],
+        fact_type=recall_fact_type,
         max_tokens=max_tokens,
         enable_trace=False,
-        request_context=request_context,
+        request_context=internal_ctx,
         tags=tags,
         tags_match=tags_match,
+        tag_groups=tag_groups,
         _connection_budget=connection_budget,
         _quiet=True,  # Suppress logging for internal operations
         include_chunks=include_chunks,

hindsight-api-slim/hindsight_api/engine/reflect/tools_schema.py

@@ -227,7 +227,12 @@ def _build_done_tool_with_directives(directive_rules: list[str]) -> dict:
     }
 
 
-def get_reflect_tools(directive_rules: list[str] | None = None) -> list[dict]:
+def get_reflect_tools(
+    directive_rules: list[str] | None = None,
+    include_mental_models: bool = True,
+    include_observations: bool = True,
+    include_recall: bool = True,
+) -> list[dict]:
     """
     Get the list of tools for the reflect agent.
 
@@ -239,16 +244,23 @@ def get_reflect_tools(directive_rules: list[str] | None = None) -> list[dict]:
     Args:
         directive_rules: Optional list of directive rule strings. If provided,
                         the done() tool will require directive compliance confirmation.
+        include_mental_models: Whether to include the search_mental_models tool.
+        include_observations: Whether to include the search_observations tool.
+        include_recall: Whether to include the recall tool.
 
     Returns:
         List of tool definitions in OpenAI format
     """
-    tools = [
-        TOOL_SEARCH_MENTAL_MODELS,
-        TOOL_SEARCH_OBSERVATIONS,
-        TOOL_RECALL,
-        TOOL_EXPAND,
-    ]
+    tools = []
+
+    if include_mental_models:
+        tools.append(TOOL_SEARCH_MENTAL_MODELS)
+    if include_observations:
+        tools.append(TOOL_SEARCH_OBSERVATIONS)
+    if include_recall:
+        tools.append(TOOL_RECALL)
+
+    tools.append(TOOL_EXPAND)
 
     # Use directive-aware done tool if directives are present
     if directive_rules:

hindsight-api-slim/hindsight_api/engine/response_models.py

@@ -8,9 +8,8 @@ API stability even if internal models change.
 
 from typing import Any
 
-from pydantic import BaseModel, ConfigDict, Field
+from pydantic import BaseModel, ConfigDict, Field, field_validator
 
-# Valid fact types for recall operations (excludes 'opinion' which is deprecated)
 VALID_RECALL_FACT_TYPES = frozenset(["world", "experience", "observation"])
 
 
@@ -20,6 +19,10 @@ class LLMToolCall(BaseModel):
     id: str = Field(description="Unique identifier for this tool call")
     name: str = Field(description="Name of the tool to call")
     arguments: dict[str, Any] = Field(description="Arguments to pass to the tool")
+    thought_signature: str | None = Field(
+        default=None,
+        description="Opaque token required by Gemini 3.1+ thinking models to preserve thought context across turns",
+    )
 
 
 class LLMToolCallResult(BaseModel):
@@ -155,6 +158,19 @@ class MemoryFact(BaseModel):
     mentioned_at: str | None = Field(None, description="ISO format date when the fact was mentioned/learned")
     document_id: str | None = Field(None, description="ID of the document this memory belongs to")
     metadata: dict[str, str] | None = Field(None, description="User-defined metadata")
+
+    @field_validator("metadata", mode="before")
+    @classmethod
+    def parse_metadata(cls, v: Any) -> dict[str, str] | None:
+        """Parse metadata from JSON string if needed (asyncpg may return JSONB as str)."""
+        if v is None:
+            return None
+        if isinstance(v, str):
+            import json
+
+            return json.loads(v)
+        return v
+
     chunk_id: str | None = Field(
         None, description="ID of the chunk this fact was extracted from (format: bank_id_document_id_chunk_index)"
     )

hindsight-api-slim/hindsight_api/engine/retain/bank_utils.py

@@ -10,32 +10,47 @@ from typing import TypedDict
 
 from pydantic import BaseModel, Field
 
+from ...config import get_config
 from ..db_utils import acquire_with_retry
 from ..memory_engine import fq_table, get_current_schema
 from ..response_models import DispositionTraits
 
 logger = logging.getLogger(__name__)
 
-# Fact types that get per-bank partial HNSW indexes, mapped to their 4-char index suffix.
-_HNSW_FACT_TYPES: dict[str, str] = {
+# Fact types that get per-bank partial vector indexes, mapped to their 4-char index suffix.
+_BANK_INDEX_FACT_TYPES: dict[str, str] = {
     "world": "worl",
     "experience": "expr",
     "observation": "obsv",
 }
 
 
-def _hnsw_index_name(ft: str, internal_id: str) -> str:
-    """Deterministic, schema-safe HNSW index name for a (bank, fact_type) pair.
+def _bank_index_name(ft: str, internal_id: str) -> str:
+    """Deterministic, schema-safe vector index name for a (bank, fact_type) pair.
 
     Uses the first 16 hex chars of internal_id (8 bytes of entropy) — unique
     enough in practice, fits comfortably within PostgreSQL's 63-char identifier limit.
     """
     uid = str(internal_id).replace("-", "")[:16]
-    return f"idx_mu_emb_{_HNSW_FACT_TYPES[ft]}_{uid}"
+    return f"idx_mu_emb_{_BANK_INDEX_FACT_TYPES[ft]}_{uid}"
 
 
-async def create_bank_hnsw_indexes(conn, bank_id: str, internal_id: str) -> None:
-    """Create per-(bank, fact_type) partial HNSW indexes for a newly created bank.
+def _vector_index_clause() -> str:
+    """Return the USING clause for vector index creation based on the configured extension."""
+    ext = get_config().vector_extension
+    if ext == "pgvectorscale":
+        return "USING diskann (embedding vector_cosine_ops) WITH (num_neighbors = 50)"
+    elif ext == "vchord":
+        return "USING vchordrq (embedding vector_l2_ops)"
+    else:  # pgvector (default)
+        return "USING hnsw (embedding vector_cosine_ops)"
+
+
+async def create_bank_vector_indexes(conn, bank_id: str, internal_id: str) -> None:
+    """Create per-(bank, fact_type) partial vector indexes for a newly created bank.
+
+    Respects the HINDSIGHT_API_VECTOR_EXTENSION config to use the appropriate
+    index type (HNSW for pgvector, DiskANN for pgvectorscale, vchordrq for vchord).
 
     Called immediately after the bank row is first inserted. Safe on empty banks
     (index build is instant). Idempotent via CREATE INDEX IF NOT EXISTS.
@@ -43,24 +58,25 @@ async def create_bank_hnsw_indexes(conn, bank_id: str, internal_id: str) -> None
     """
     table = fq_table("memory_units")
     escaped = bank_id.replace("'", "''")
-    for ft in _HNSW_FACT_TYPES:
-        idx = _hnsw_index_name(ft, internal_id)
+    using_clause = _vector_index_clause()
+    for ft in _BANK_INDEX_FACT_TYPES:
+        idx = _bank_index_name(ft, internal_id)
         await conn.execute(
             f"CREATE INDEX IF NOT EXISTS {idx} "
-            f"ON {table} USING hnsw (embedding vector_cosine_ops) "
+            f"ON {table} {using_clause} "
             f"WHERE fact_type = '{ft}' AND bank_id = '{escaped}'"
         )
 
 
-async def drop_bank_hnsw_indexes(conn, internal_id: str) -> None:
-    """Drop per-(bank, fact_type) partial HNSW indexes for a bank being deleted.
+async def drop_bank_vector_indexes(conn, internal_id: str) -> None:
+    """Drop per-(bank, fact_type) partial vector indexes for a bank being deleted.
 
     Called before the bank row is deleted so internal_id is still known.
     Idempotent via DROP INDEX IF EXISTS.
     """
     schema = get_current_schema()
-    for ft in _HNSW_FACT_TYPES:
-        idx = _hnsw_index_name(ft, internal_id)
+    for ft in _BANK_INDEX_FACT_TYPES:
+        idx = _bank_index_name(ft, internal_id)
         await conn.execute(f"DROP INDEX IF EXISTS {schema}.{idx}")
 
 
@@ -97,6 +113,22 @@ async def get_bank_profile(pool, bank_id: str) -> BankProfile:
     Returns:
         BankProfile with name, typed DispositionTraits, and mission
     """
+    profile, _ = await get_or_create_bank_profile(pool, bank_id)
+    return profile
+
+
+async def get_or_create_bank_profile(pool, bank_id: str) -> tuple[BankProfile, bool]:
+    """
+    Get bank profile, auto-creating with defaults if it doesn't exist.
+
+    Same as get_bank_profile, but also returns a flag indicating whether the
+    bank was freshly created on this call. Used by the memory engine to apply
+    the HINDSIGHT_API_DEFAULT_BANK_TEMPLATE hook on first bank creation.
+
+    Returns:
+        Tuple of (BankProfile, created) where created is True if the bank
+        did not exist before this call.
+    """
     async with acquire_with_retry(pool) as conn:
         # Try to get existing bank
         row = await conn.fetchrow(
@@ -113,15 +145,18 @@ async def get_bank_profile(pool, bank_id: str) -> BankProfile:
             if isinstance(disposition_data, str):
                 disposition_data = json.loads(disposition_data)
 
-            return BankProfile(
-                name=row["name"],
-                disposition=DispositionTraits(**disposition_data),
-                mission=row["mission"] or "",
+            return (
+                BankProfile(
+                    name=row["name"],
+                    disposition=DispositionTraits(**disposition_data),
+                    mission=row["mission"] or "",
+                ),
+                False,
             )
 
         # Bank doesn't exist, create with defaults.
         # Generate internal_id here so we control the value and can use it
-        # immediately for HNSW index creation without a RETURNING round-trip.
+        # immediately for vector index creation without a RETURNING round-trip.
         internal_id = uuid.uuid4()
         inserted = await conn.fetchval(
             f"""
@@ -137,11 +172,15 @@ async def get_bank_profile(pool, bank_id: str) -> BankProfile:
             internal_id,
         )
 
-        if inserted:
-            # Fresh insert — create per-bank HNSW indexes (instant on empty bank)
-            await create_bank_hnsw_indexes(conn, bank_id, str(internal_id))
+        created = inserted is not None
+        if created:
+            # Fresh insert — create per-bank vector indexes (instant on empty bank)
+            await create_bank_vector_indexes(conn, bank_id, str(internal_id))
 
-        return BankProfile(name=bank_id, disposition=DispositionTraits(**DEFAULT_DISPOSITION), mission="")
+        return (
+            BankProfile(name=bank_id, disposition=DispositionTraits(**DEFAULT_DISPOSITION), mission=""),
+            created,
+        )
 
 
 async def update_bank_disposition(pool, bank_id: str, disposition: dict[str, int]) -> None:

hindsight-api-slim/hindsight_api/engine/retain/chunk_storage.py

@@ -4,7 +4,9 @@ Chunk storage for retain pipeline.
 Handles storage of document chunks in the database.
 """
 
+import hashlib
 import logging
+from dataclasses import dataclass
 
 from ..memory_engine import fq_table
 from .types import ChunkMetadata
@@ -12,6 +14,61 @@ from .types import ChunkMetadata
 logger = logging.getLogger(__name__)
 
 
+def compute_chunk_hash(chunk_text: str) -> str:
+    """Compute SHA256 hash of chunk text for delta comparison."""
+    return hashlib.sha256(chunk_text.encode()).hexdigest()
+
+
+@dataclass
+class ExistingChunk:
+    """Represents a chunk already stored in the database."""
+
+    chunk_id: str
+    chunk_index: int
+    content_hash: str | None
+
+
+async def load_existing_chunks(conn, bank_id: str, document_id: str) -> list[ExistingChunk]:
+    """
+    Load existing chunk metadata for a document.
+
+    Returns list of ExistingChunk with chunk_id, chunk_index, and content_hash.
+    """
+    rows = await conn.fetch(
+        f"""
+        SELECT chunk_id, chunk_index, content_hash
+        FROM {fq_table("chunks")}
+        WHERE document_id = $1 AND bank_id = $2
+        ORDER BY chunk_index
+        """,
+        document_id,
+        bank_id,
+    )
+    return [
+        ExistingChunk(
+            chunk_id=row["chunk_id"],
+            chunk_index=row["chunk_index"],
+            content_hash=row["content_hash"],
+        )
+        for row in rows
+    ]
+
+
+async def delete_chunks_by_ids(conn, chunk_ids: list[str]) -> None:
+    """
+    Delete specific chunks by their IDs.
+
+    This cascades to memory_units (via FK with CASCADE delete)
+    and their links.
+    """
+    if not chunk_ids:
+        return
+    await conn.execute(
+        f"DELETE FROM {fq_table('chunks')} WHERE chunk_id = ANY($1::text[])",
+        chunk_ids,
+    )
+
+
 async def store_chunks_batch(conn, bank_id: str, document_id: str, chunks: list[ChunkMetadata]) -> dict[int, str]:
     """
     Store document chunks in the database.
@@ -32,6 +89,7 @@ async def store_chunks_batch(conn, bank_id: str, document_id: str, chunks: list[
     chunk_ids = []
     chunk_texts = []
     chunk_indices = []
+    content_hashes = []
     chunk_id_map = {}
 
     for chunk in chunks:
@@ -39,19 +97,30 @@ async def store_chunks_batch(conn, bank_id: str, document_id: str, chunks: list[
         chunk_ids.append(chunk_id)
         chunk_texts.append(chunk.chunk_text)
         chunk_indices.append(chunk.chunk_index)
+        content_hashes.append(compute_chunk_hash(chunk.chunk_text))
         chunk_id_map[chunk.chunk_index] = chunk_id
 
-    # Batch insert all chunks
+    # Batch upsert all chunks. ON CONFLICT makes this idempotent: re-submitting
+    # a retain under the same document_id (the pattern in vectorize-io/hindsight#977)
+    # may produce chunk_ids that already exist when upstream cascade-delete or
+    # delta-retain paths don't run (or race with a concurrent task). Overwriting
+    # is the correct behavior per the document_id grouping semantics — the caller
+    # intends this chunk to hold the latest content at that (document_id, index).
     await conn.execute(
         f"""
-        INSERT INTO {fq_table("chunks")} (chunk_id, document_id, bank_id, chunk_text, chunk_index)
-        SELECT * FROM unnest($1::text[], $2::text[], $3::text[], $4::text[], $5::integer[])
+        INSERT INTO {fq_table("chunks")} (chunk_id, document_id, bank_id, chunk_text, chunk_index, content_hash)
+        SELECT * FROM unnest($1::text[], $2::text[], $3::text[], $4::text[], $5::integer[], $6::text[])
+        ON CONFLICT (chunk_id) DO UPDATE SET
+            chunk_text = EXCLUDED.chunk_text,
+            chunk_index = EXCLUDED.chunk_index,
+            content_hash = EXCLUDED.content_hash
         """,
         chunk_ids,
         [document_id] * len(chunk_texts),
         [bank_id] * len(chunk_texts),
         chunk_texts,
         chunk_indices,
+        content_hashes,
     )
 
     return chunk_id_map

hindsight-api-slim/hindsight_api/engine/retain/embedding_utils.py

@@ -47,6 +47,16 @@ async def generate_embeddings_batch(embeddings_backend, texts: list[str]) -> lis
             embeddings_backend.encode,
             texts,
         )
-        return embeddings
     except Exception as e:
         raise Exception(f"Failed to generate batch embeddings: {str(e)}")
+
+    # Guarantee 1:1 alignment with input texts. A silent length mismatch here
+    # propagates downstream as zip() drops items, eventually surfacing as an
+    # IndexError in retain mapping (see issue #1037).
+    if len(embeddings) != len(texts):
+        raise RuntimeError(
+            f"Embeddings backend returned {len(embeddings)} vectors for {len(texts)} input texts; "
+            "expected exact 1:1 alignment"
+        )
+
+    return embeddings

hindsight-api-slim/hindsight_api/engine/retain/entity_processing.py

@@ -12,61 +12,27 @@ from .types import EntityLink, ProcessedFact
 logger = logging.getLogger(__name__)
 
 
-async def process_entities_batch(
-    entity_resolver,
-    conn,
-    bank_id: str,
-    unit_ids: list[str],
+def _prepare_facts_for_entity_processing(
     facts: list[ProcessedFact],
-    log_buffer: list[str] = None,
-    user_entities_per_content: dict[int, list[dict]] = None,
-    entity_labels: list | None = None,
-) -> list[EntityLink]:
+    user_entities_per_content: dict[int, list[dict]] | None = None,
+) -> tuple[list[str], list, list[list[dict]]]:
     """
-    Process entities for all facts and create entity links.
-
-    This function:
-    1. Extracts entity mentions from fact texts
-    2. Merges user-provided entities with LLM-extracted entities
-    3. Resolves entity names to canonical entities
-    4. Creates entity records in the database
-    5. Returns entity links ready for insertion
-
-    Args:
-        entity_resolver: EntityResolver instance for entity resolution
-        conn: Database connection
-        bank_id: Bank identifier
-        unit_ids: List of unit IDs (same length as facts)
-        facts: List of ProcessedFact objects
-        log_buffer: Optional buffer for detailed logging
-        user_entities_per_content: Dict mapping content_index to list of user-provided entities
+    Extract fact texts, dates, and merged entity lists from ProcessedFact objects.
 
     Returns:
-        List of EntityLink objects for batch insertion
+        Tuple of (fact_texts, fact_dates, entities_per_fact)
     """
-    if not unit_ids or not facts:
-        return []
-
-    if len(unit_ids) != len(facts):
-        raise ValueError(f"Mismatch between unit_ids ({len(unit_ids)}) and facts ({len(facts)})")
-
     user_entities_per_content = user_entities_per_content or {}
 
-    # Extract data for link_utils function
     fact_texts = [fact.fact_text for fact in facts]
-    # Use occurred_start if available, otherwise use mentioned_at for entity timestamps
     fact_dates = [fact.occurred_start if fact.occurred_start is not None else fact.mentioned_at for fact in facts]
 
-    # Convert EntityRef objects to dict format and merge with user-provided entities
     entities_per_fact = []
     for fact in facts:
-        # Start with LLM-extracted entities
         llm_entities = [{"text": entity.name, "type": "CONCEPT"} for entity in (fact.entities or [])]
 
-        # Get user entities for this content (use content_index from fact)
         user_entities = user_entities_per_content.get(fact.content_index, [])
 
-        # Merge with case-insensitive deduplication
         seen_texts = {e["text"].lower() for e in llm_entities}
         for user_entity in user_entities:
             if user_entity["text"].lower() not in seen_texts:
@@ -80,8 +46,48 @@ async def process_entities_batch(
 
         entities_per_fact.append(llm_entities)
 
-    # Use existing link_utils function for entity processing
-    entity_links = await link_utils.extract_entities_batch_optimized(
+    return fact_texts, fact_dates, entities_per_fact
+
+
+async def resolve_entities(
+    entity_resolver,
+    conn,
+    bank_id: str,
+    unit_ids: list[str],
+    facts: list[ProcessedFact],
+    log_buffer: list[str] = None,
+    user_entities_per_content: dict[int, list[dict]] = None,
+    entity_labels: list | None = None,
+) -> tuple[list[str], list[tuple], dict[str, list[str]]]:
+    """
+    Phase 1: Resolve entity names to canonical IDs (read-heavy).
+
+    Should be called on a SEPARATE connection OUTSIDE the main write transaction
+    to avoid holding the transaction open during expensive trigram scans.
+
+    Args:
+        entity_resolver: EntityResolver instance
+        conn: Database connection (separate from the main write transaction)
+        bank_id: Bank identifier
+        unit_ids: Placeholder unit IDs (used only for grouping)
+        facts: List of ProcessedFact objects
+        log_buffer: Optional buffer for detailed logging
+        user_entities_per_content: Dict mapping content_index to user-provided entities
+        entity_labels: Optional entity label taxonomy
+
+    Returns:
+        Tuple of (resolved_entity_ids, entity_to_unit, unit_to_entity_ids)
+        to pass to build_entity_links().
+    """
+    if not unit_ids or not facts:
+        return [], [], {}
+
+    if len(unit_ids) != len(facts):
+        raise ValueError(f"Mismatch between unit_ids ({len(unit_ids)}) and facts ({len(facts)})")
+
+    fact_texts, fact_dates, entities_per_fact = _prepare_facts_for_entity_processing(facts, user_entities_per_content)
+
+    return await link_utils.resolve_entities_only(
         entity_resolver,
         conn,
         bank_id,
@@ -90,22 +96,67 @@ async def process_entities_batch(
         "",  # context (not used in current implementation)
         fact_dates,
         entities_per_fact,
-        log_buffer,  # Pass log_buffer for detailed logging
+        log_buffer,
         entity_labels=entity_labels,
     )
 
-    return entity_links
+
+async def build_entity_links(
+    entity_resolver,
+    conn,
+    bank_id: str,
+    unit_ids: list[str],
+    resolved_entity_ids: list[str],
+    entity_to_unit: list[tuple],
+    unit_to_entity_ids: dict[str, list[str]],
+    log_buffer: list[str] = None,
+    skip_unit_entities_insert: bool = False,
+) -> list[EntityLink]:
+    """
+    Build entity links for UI graph visualization.
+
+    Queries unit_entities to find shared entities between new and existing units,
+    then generates EntityLink objects. When called from Phase 3 (post-transaction),
+    set skip_unit_entities_insert=True since unit_entities were already inserted
+    in Phase 2.
+
+    Args:
+        entity_resolver: EntityResolver instance
+        conn: Database connection
+        bank_id: Bank identifier
+        unit_ids: Actual unit IDs (must already be inserted in the DB)
+        resolved_entity_ids: From resolve_entities()
+        entity_to_unit: From resolve_entities()
+        unit_to_entity_ids: From resolve_entities()
+        log_buffer: Optional buffer for detailed logging
+        skip_unit_entities_insert: Skip unit_entities INSERT (already done in Phase 2)
+
+    Returns:
+        List of EntityLink objects for batch insertion
+    """
+    return await link_utils.build_entity_links_from_resolved(
+        entity_resolver,
+        conn,
+        bank_id,
+        unit_ids,
+        resolved_entity_ids,
+        entity_to_unit,
+        unit_to_entity_ids,
+        log_buffer,
+        skip_unit_entities_insert=skip_unit_entities_insert,
+    )
 
 
-async def insert_entity_links_batch(conn, entity_links: list[EntityLink]) -> None:
+async def insert_entity_links_batch(conn, entity_links: list[EntityLink], bank_id: str) -> None:
     """
     Insert entity links in batch.
 
     Args:
         conn: Database connection
         entity_links: List of EntityLink objects
+        bank_id: Bank identifier (stored directly on memory_links for fast filtering)
     """
     if not entity_links:
         return
 
-    await link_utils.insert_entity_links_batch(conn, entity_links)
+    await link_utils.insert_entity_links_batch(conn, entity_links, bank_id)

hindsight-api-slim/hindsight_api/engine/retain/fact_extraction.py

@@ -87,7 +87,7 @@ class Fact(BaseModel):
 
     # Required fields
     fact: str = Field(description="Combined fact text: what | when | where | who | why")
-    fact_type: Literal["world", "experience", "opinion"] = Field(description="Perspective: world/experience/opinion")
+    fact_type: Literal["world", "experience"] = Field(description="Perspective: world/experience")
 
     # Optional temporal fields
     occurred_start: str | None = None
@@ -159,7 +159,9 @@ class ExtractedFact(BaseModel):
     fact_kind: str = Field(default="conversation", description="'event' or 'conversation'")
     occurred_start: str | None = Field(default=None, description="ISO timestamp for events")
     occurred_end: str | None = Field(default=None, description="ISO timestamp for event end")
-    fact_type: Literal["world", "assistant"] = Field(description="'world' or 'assistant'")
+    fact_type: Literal["world", "assistant"] = Field(
+        description="'world' = objective/external facts. 'assistant' = first-person actions, experiences, or observations by the speaker."
+    )
     entities: list[Entity] | None = Field(default=None, description="People, places, concepts")
     causal_relations: list[FactCausalRelation] | None = Field(
         default=None, description="Links to previous facts (target_index < this fact's index)"
@@ -261,7 +263,7 @@ class ExtractedFactVerbose(BaseModel):
     )
 
     fact_type: Literal["world", "assistant"] = Field(
-        description="'world' = about the user/others (background, experiences). 'assistant' = experience with the assistant."
+        description="'world' = objective/external facts about other people, events, general knowledge. 'assistant' = first-person actions, experiences, or observations by the speaker (e.g., 'I changed X', 'I discovered Y')."
     )
 
     entities: list[Entity] | None = Field(
@@ -332,6 +334,45 @@ class FactExtractionResponseNoCausal(BaseModel):
     facts: list[ExtractedFactNoCausal] = Field(description="List of extracted factual statements")
 
 
+class VerbatimExtractedFact(BaseModel):
+    """
+    Schema for verbatim extraction mode.
+
+    Omits 'what' entirely — the original chunk text is used as fact_text in code.
+    The LLM only extracts metadata: entities, temporal info, location, people.
+    """
+
+    model_config = ConfigDict(
+        json_schema_mode="validation",
+        json_schema_extra={"required": ["when", "where", "who", "fact_type"]},
+    )
+
+    when: str = Field(description="When it happened. 'N/A' if unknown.")
+    where: str = Field(description="Location if relevant. 'N/A' if none.")
+    who: str = Field(description="People involved with relationships. 'N/A' if general.")
+
+    fact_kind: str = Field(default="conversation", description="'event' or 'conversation'")
+    occurred_start: str | None = Field(default=None, description="ISO timestamp for events")
+    occurred_end: str | None = Field(default=None, description="ISO timestamp for event end")
+    fact_type: Literal["world", "assistant"] = Field(
+        description="'world' = objective/external facts. 'assistant' = first-person actions, experiences, or observations by the speaker."
+    )
+    entities: list[Entity] | None = Field(default=None, description="People, places, concepts")
+
+    @field_validator("entities", mode="before")
+    @classmethod
+    def ensure_entities_list(cls, v):
+        if v is None:
+            return []
+        return v
+
+
+class VerbatimFactExtractionResponse(BaseModel):
+    """Response for verbatim extraction mode (one entry per chunk, no fact text)."""
+
+    facts: list[VerbatimExtractedFact] = Field(description="List of metadata entries (one per chunk)")
+
+
 def chunk_text(text: str, max_chars: int) -> list[str]:
     """
     Split text into chunks, preserving conversation structure when possible.
@@ -462,8 +503,8 @@ fact_kind:
 - "conversation": Ongoing state, preference, trait (no dates)
 
 fact_type:
-- "world": About user's life, other people, external events
-- "assistant": Interactions with assistant (requests, recommendations)
+- "world": About other people, external events, general knowledge, objective facts
+- "assistant": First-person actions, experiences, or observations by the speaker/author (e.g., "I changed X", "I discovered Y", "I debugged Z"). Also includes interactions with the user (requests, recommendations). If the narrator describes something they did, tried, learned, or decided — use "assistant".
 
 ══════════════════════════════════════════════════════════════════════════
 TEMPORAL HANDLING
@@ -552,13 +593,34 @@ CUSTOM_FACT_EXTRACTION_PROMPT = _BASE_FACT_EXTRACTION_PROMPT.format(
     examples="",  # No examples for custom mode
 )
 
+# Verbatim mode: preserve the original text exactly, but still extract metadata
+_VERBATIM_GUIDELINES = """══════════════════════════════════════════════════════════════════════════
+VERBATIM MODE — Extract metadata only
+══════════════════════════════════════════════════════════════════════════
+
+The original text will be stored as-is in code. Your ONLY job is to extract metadata.
+
+RULES:
+- Produce EXACTLY ONE entry per input chunk.
+- DO NOT include a "what" field — it is not part of the output schema.
+- Extract all entities (people, places, organizations, objects, concepts).
+- Extract temporal information (occurred_start, occurred_end, fact_kind, when).
+- Extract location (where) and people (who).
+- fact_type: use "world" unless the content is clearly an interaction with the assistant."""
+
+VERBATIM_FACT_EXTRACTION_PROMPT = _BASE_FACT_EXTRACTION_PROMPT.format(
+    retain_mission_section="{retain_mission_section}",
+    extraction_guidelines=_VERBATIM_GUIDELINES,
+    examples="",
+)
+
 
 # Verbose extraction prompt - detailed, comprehensive facts (legacy mode)
 VERBOSE_FACT_EXTRACTION_PROMPT = """Extract facts from text into structured format with FIVE required dimensions - BE EXTREMELY DETAILED.
 
 LANGUAGE: MANDATORY — Detect the language of the input text and produce ALL output in that EXACT same language. You are STRICTLY FORBIDDEN from translating or switching to any other language. Every single word of your output must be in the same language as the input. Do NOT output in a different language under any circumstance.
 
-══════════════════════════════════════════════════════════════════════════
+{retain_mission_section}══════════════════════════════════════════════════════════════════════════
 FACT FORMAT - ALL FIVE DIMENSIONS REQUIRED - MAXIMUM VERBOSITY
 ══════════════════════════════════════════════════════════════════════════
 
@@ -769,7 +831,13 @@ def _build_extraction_prompt_and_schema(config) -> tuple[str, type]:
                 custom_instructions=config.retain_custom_instructions,
             )
     elif extraction_mode == "verbose":
-        prompt = VERBOSE_FACT_EXTRACTION_PROMPT
+        prompt = VERBOSE_FACT_EXTRACTION_PROMPT.format(
+            retain_mission_section=retain_mission_section,
+        )
+    elif extraction_mode == "verbatim":
+        prompt = VERBATIM_FACT_EXTRACTION_PROMPT.format(
+            retain_mission_section=retain_mission_section,
+        )
     else:
         base_prompt = CONCISE_FACT_EXTRACTION_PROMPT
         prompt = base_prompt.format(
@@ -777,7 +845,11 @@ def _build_extraction_prompt_and_schema(config) -> tuple[str, type]:
         )
 
     # Add causal relationships section if enabled
-    if extract_causal_links:
+    # Verbatim mode never uses causal relations (no fact text to relate causally)
+    if extraction_mode == "verbatim":
+        base_fact_class = VerbatimExtractedFact
+        base_response_class = VerbatimFactExtractionResponse
+    elif extract_causal_links:
         prompt = prompt + CAUSAL_RELATIONSHIPS_SECTION
         base_fact_class = ExtractedFactVerbose if extraction_mode == "verbose" else ExtractedFact
         base_response_class = FactExtractionResponseVerbose if extraction_mode == "verbose" else FactExtractionResponse
@@ -837,6 +909,7 @@ def _build_user_message(
     event_date: datetime | None,
     context: str,
     metadata: dict[str, str] | None = None,
+    agent_name: str | None = None,
 ) -> str:
     """Build user message for fact extraction."""
     from .orchestrator import parse_datetime_flexible
@@ -855,11 +928,15 @@ def _build_user_message(
         metadata_lines = "\n".join(f"  {k}: {v}" for k, v in metadata.items())
         metadata_section = f"\nMetadata:\n{metadata_lines}"
 
+    narrator_section = ""
+    if agent_name:
+        narrator_section = f'\nNarrator: {agent_name} (AI agent — first-person statements like "I did X" are the agent\'s own actions; classify as "assistant")'
+
     return f"""Extract facts from the following text chunk.
 
 Chunk: {chunk_index + 1}/{total_chunks}
 Event Date: {event_date_str}
-Context: {sanitized_context}{metadata_section}
+Context: {sanitized_context}{metadata_section}{narrator_section}
 
 Text:
 {sanitized_chunk}"""
@@ -923,7 +1000,7 @@ async def _extract_facts_from_chunk(
     extract_causal_links = config.retain_extract_causal_links
 
     # Build user message using helper function
-    user_message = _build_user_message(chunk, chunk_index, total_chunks, event_date, context, metadata)
+    user_message = _build_user_message(chunk, chunk_index, total_chunks, event_date, context, metadata, agent_name)
 
     # Retry logic for JSON validation errors
     # Use retain-specific overrides if set, otherwise fall back to global LLM config
@@ -983,7 +1060,7 @@ async def _extract_facts_from_chunk(
                     f"LLM response missing 'facts' field or returned empty list. "
                     f"Response: {extraction_response_json}. "
                     f"Input: "
-                    f"date: {event_date.isoformat()}, "
+                    f"date: {event_date.isoformat() if event_date else 'unset'}, "
                     f"context: {context if context else 'none'}, "
                     f"text: {chunk}"
                 )
@@ -1012,33 +1089,21 @@ async def _extract_facts_from_chunk(
                 if not what:
                     what = get_value("factual_core")
                 if not what:
-                    logger.warning(f"Skipping fact {i}: missing 'what' field")
-                    continue
+                    # In verbatim mode, 'what' is intentionally absent — text is backfilled from chunk
+                    if extraction_mode != "verbatim":
+                        logger.warning(f"Skipping fact {i}: missing 'what' field")
+                        continue
 
-                # Critical field: fact_type
-                # LLM uses "assistant" but we convert to "experience" for storage
-                original_fact_type = llm_fact.get("fact_type")
-                fact_type = original_fact_type
-
-                # Convert "assistant" → "experience" for storage
-                if fact_type == "assistant":
+                # Critical field: fact_type — "assistant" maps to "experience", everything else is "world".
+                # If fact_type is unexpected, fall back to fact_kind before defaulting to "world".
+                raw_fact_type = llm_fact.get("fact_type")
+                if raw_fact_type == "assistant":
                     fact_type = "experience"
-
-                # Validate fact_type (after conversion)
-                if fact_type not in ["world", "experience", "opinion"]:
-                    # Try to fix common mistakes - check if they swapped fact_type and fact_kind
-                    fact_kind = llm_fact.get("fact_kind")
-                    if fact_kind == "assistant":
-                        fact_type = "experience"
-                    elif fact_kind in ["world", "experience", "opinion"]:
-                        fact_type = fact_kind
-                    else:
-                        # Default to 'world' if we can't determine
-                        fact_type = "world"
-                        logger.warning(
-                            f"Fact {i}: defaulting to fact_type='world' "
-                            f"(original fact_type={original_fact_type!r}, fact_kind={fact_kind!r})"
-                        )
+                elif raw_fact_type == "world":
+                    fact_type = "world"
+                else:
+                    raw_fact_kind = llm_fact.get("fact_kind")
+                    fact_type = "experience" if raw_fact_kind == "assistant" else "world"
 
                 # Get fact_kind for temporal handling (but don't store it)
                 fact_kind = llm_fact.get("fact_kind", "conversation")
@@ -1046,19 +1111,23 @@ async def _extract_facts_from_chunk(
                     fact_kind = "conversation"
 
                 # Build combined fact text from the 4 dimensions: what | when | who | why
+                # In verbatim mode, leave combined_text empty — _collapse_to_verbatim backfills it
                 fact_data = {}
-                combined_parts = [what]
+                if extraction_mode == "verbatim":
+                    combined_text = ""
+                else:
+                    combined_parts = [what]
 
-                if when:
-                    combined_parts.append(f"When: {when}")
+                    if when:
+                        combined_parts.append(f"When: {when}")
 
-                if who:
-                    combined_parts.append(f"Involving: {who}")
+                    if who:
+                        combined_parts.append(f"Involving: {who}")
 
-                if why:
-                    combined_parts.append(why)
+                    if why:
+                        combined_parts.append(why)
 
-                combined_text = " | ".join(combined_parts)
+                    combined_text = " | ".join(combined_parts)
 
                 # Add temporal fields
                 # For events: occurred_start/occurred_end (when the event happened)
@@ -1400,28 +1469,76 @@ async def extract_facts_from_text(
             f"chunk_size={config.retain_chunk_size:,}) - starting parallel LLM extraction"
         )
 
-    tasks = [
-        _extract_facts_with_auto_split(
-            chunk=chunk,
-            chunk_index=i,
-            total_chunks=len(chunks),
-            event_date=event_date,
-            context=context,
-            llm_config=llm_config,
-            config=config,
-            agent_name=agent_name,
-            metadata=metadata,
-        )
-        for i, chunk in enumerate(chunks)
-    ]
-    chunk_results = await asyncio.gather(*tasks)
+    # Per-chunk retry wrapper: each chunk gets up to MAX_CHUNK_RETRIES attempts.
+    # This handles transient LLM failures (timeouts, rate limits, malformed responses)
+    # without discarding the entire batch. If a chunk still fails after all retries,
+    # the ENTIRE retain fails — we do not accept partial extraction.
+    MAX_CHUNK_RETRIES = 3
+    CHUNK_RETRY_BASE_DELAY = 2.0  # seconds, doubles each retry
+
+    async def _extract_chunk_with_retry(chunk: str, chunk_index: int) -> tuple:
+        """Extract facts from a single chunk with retries on failure."""
+        last_exception = None
+        for attempt in range(MAX_CHUNK_RETRIES):
+            try:
+                return await _extract_facts_with_auto_split(
+                    chunk=chunk,
+                    chunk_index=chunk_index,
+                    total_chunks=len(chunks),
+                    event_date=event_date,
+                    context=context,
+                    llm_config=llm_config,
+                    config=config,
+                    agent_name=agent_name,
+                    metadata=metadata,
+                )
+            except Exception as e:
+                last_exception = e
+                if attempt < MAX_CHUNK_RETRIES - 1:
+                    delay = CHUNK_RETRY_BASE_DELAY * (2**attempt)
+                    logger.warning(
+                        f"Chunk {chunk_index}/{len(chunks)} extraction failed "
+                        f"(attempt {attempt + 1}/{MAX_CHUNK_RETRIES}): "
+                        f"{type(e).__name__}. Retrying in {delay:.0f}s..."
+                    )
+                    await asyncio.sleep(delay)
+                else:
+                    logger.error(
+                        f"Chunk {chunk_index}/{len(chunks)} extraction failed after "
+                        f"{MAX_CHUNK_RETRIES} attempts: {type(e).__name__}: {e}"
+                    )
+        raise last_exception
+
+    tasks = [_extract_chunk_with_retry(chunk, i) for i, chunk in enumerate(chunks)]
+
+    # return_exceptions=True so we can collect all results even if some chunks
+    # exhausted their retries. We check for failures below and fail the retain
+    # if ANY chunk could not be extracted — partial extraction is not acceptable.
+    chunk_results = await asyncio.gather(*tasks, return_exceptions=True)
+
     all_facts = []
     chunk_metadata = []  # [(chunk_text, fact_count), ...]
     total_usage = TokenUsage()
-    for chunk, (chunk_facts, chunk_usage) in zip(chunks, chunk_results):
+    failed_chunks = []
+    for i, (chunk, result) in enumerate(zip(chunks, chunk_results)):
+        if isinstance(result, Exception):
+            failed_chunks.append((i, result))
+            continue
+        chunk_facts, chunk_usage = result
         all_facts.extend(chunk_facts)
         chunk_metadata.append((chunk, len(chunk_facts)))
         total_usage = total_usage + chunk_usage
+
+    if failed_chunks:
+        # Fail the entire retain — partial extraction is not acceptable.
+        # All successfully extracted facts are discarded because the transaction
+        # hasn't committed yet. The worker poller will retry the entire task.
+        failed_summary = ", ".join(f"chunk {idx}: {type(err).__name__}" for idx, err in failed_chunks[:5])
+        raise RuntimeError(
+            f"Fact extraction failed: {len(failed_chunks)}/{len(chunks)} chunks failed "
+            f"after {MAX_CHUNK_RETRIES} retries each. First failures: {failed_summary}"
+        )
+
     return all_facts, chunk_metadata, total_usage
 
 
@@ -1520,7 +1637,13 @@ async def extract_facts_from_contents_batch_api(
 
             # Build user message using helper function
             user_message = _build_user_message(
-                chunk, chunk_index_in_content, len(chunks), item.event_date, item.context, item.metadata or None
+                chunk,
+                chunk_index_in_content,
+                len(chunks),
+                item.event_date,
+                item.context,
+                item.metadata or None,
+                agent_name,
             )
 
             # Build request body using helper function
@@ -1682,23 +1805,17 @@ async def extract_facts_from_contents_batch_api(
             who = get_value("who")
             why = get_value("why")
 
-            # Critical field: fact_type
-            original_fact_type = llm_fact.get("fact_type")
-            fact_type = original_fact_type
-
-            # Convert "assistant" → "experience"
-            if fact_type == "assistant":
+            # Critical field: fact_type — only "assistant" maps to "experience", everything else is "world"
+            # Critical field: fact_type — "assistant" maps to "experience", everything else is "world".
+            # If fact_type is unexpected, fall back to fact_kind before defaulting to "world".
+            raw_fact_type = llm_fact.get("fact_type")
+            if raw_fact_type == "assistant":
                 fact_type = "experience"
-
-            # Validate fact_type
-            if fact_type not in ["world", "experience", "opinion"]:
-                fact_kind = llm_fact.get("fact_kind")
-                if fact_kind == "assistant":
-                    fact_type = "experience"
-                elif fact_kind in ["world", "experience", "opinion"]:
-                    fact_type = fact_kind
-                else:
-                    fact_type = "world"
+            elif raw_fact_type == "world":
+                fact_type = "world"
+            else:
+                raw_fact_kind = llm_fact.get("fact_kind")
+                fact_type = "experience" if raw_fact_kind == "assistant" else "world"
 
             # Build combined fact text
             combined_parts = [what]
@@ -1889,6 +2006,52 @@ async def extract_facts_from_contents_batch_api(
     return extracted_facts, chunks_metadata, total_usage
 
 
+def _extract_facts_chunks(
+    contents: list[RetainContent],
+    config,
+) -> tuple[list[ExtractedFactType], list[ChunkMetadata], TokenUsage]:
+    """
+    chunks mode: no LLM call, no entity extraction.
+
+    Each chunk becomes one memory unit with the raw text as fact_text.
+    User-provided entities from RetainContent.entities are picked up downstream
+    by entity_processing.py — they are the sole source of entity data in this mode.
+    """
+    extracted_facts: list[ExtractedFactType] = []
+    chunks_metadata: list[ChunkMetadata] = []
+    global_chunk_idx = 0
+
+    for content_index, content in enumerate(contents):
+        chunks = chunk_text(content.content, config.retain_chunk_size)
+        for chunk in chunks:
+            chunks_metadata.append(
+                ChunkMetadata(
+                    chunk_text=chunk,
+                    fact_count=1,
+                    content_index=content_index,
+                    chunk_index=global_chunk_idx,
+                )
+            )
+            extracted_facts.append(
+                ExtractedFactType(
+                    fact_text=chunk,
+                    fact_type="world",
+                    entities=[],
+                    content_index=content_index,
+                    chunk_index=global_chunk_idx,
+                    context=content.context,
+                    mentioned_at=content.event_date,
+                    metadata=content.metadata,
+                    tags=content.tags,
+                    observation_scopes=content.observation_scopes,
+                )
+            )
+            global_chunk_idx += 1
+
+    _add_temporal_offsets(extracted_facts, contents)
+    return extracted_facts, chunks_metadata, TokenUsage()
+
+
 async def extract_facts_from_contents(
     contents: list[RetainContent],
     llm_config,
@@ -1924,6 +2087,11 @@ async def extract_facts_from_contents(
     if not contents:
         return [], [], TokenUsage()
 
+    # chunks mode: skip LLM entirely, store each chunk as-is
+    # Must come before the batch-API check so no LLM queue/locks are acquired
+    if config.retain_extraction_mode == "chunks":
+        return _extract_facts_chunks(contents, config)
+
     # Route to batch API if enabled
     if config.retain_batch_enabled:
         return await extract_facts_from_contents_batch_api(
@@ -1946,8 +2114,9 @@ async def extract_facts_from_contents(
         )
         fact_extraction_tasks.append(task)
 
-    # Step 2: Wait for all fact extractions to complete
-    all_fact_results = await asyncio.gather(*fact_extraction_tasks)
+    # Step 2: Wait for all fact extractions to complete.
+    # Use return_exceptions=True so one content item failure doesn't discard the rest.
+    all_fact_results = await asyncio.gather(*fact_extraction_tasks, return_exceptions=True)
 
     # Step 3: Flatten and convert to typed objects
     extracted_facts: list[ExtractedFactType] = []
@@ -1957,9 +2126,16 @@ async def extract_facts_from_contents(
     global_chunk_idx = 0
     global_fact_idx = 0
 
-    for content_index, (content, (facts_from_llm, chunks_from_llm, content_usage)) in enumerate(
-        zip(contents, all_fact_results)
-    ):
+    # Filter out failed content items
+    valid_results = []
+    for content, result in zip(contents, all_fact_results):
+        if isinstance(result, Exception):
+            logger.warning(f"Content extraction failed (skipping): {type(result).__name__}: {result}")
+            valid_results.append((content, ([], [], TokenUsage())))
+        else:
+            valid_results.append((content, result))
+
+    for content_index, (content, (facts_from_llm, chunks_from_llm, content_usage)) in enumerate(valid_results):
         total_usage = total_usage + content_usage
         chunk_start_idx = global_chunk_idx
 
@@ -2013,15 +2189,46 @@ async def extract_facts_from_contents(
                     global_fact_idx += 1
                     fact_idx_in_content += 1
 
-    # Step 4: Add time offsets to preserve ordering within each content
+    # Step 4: For verbatim mode, collapse to one fact per chunk with original text
+    if config.retain_extraction_mode == "verbatim":
+        extracted_facts = _collapse_to_verbatim(extracted_facts, chunks_metadata)
+
+    # Step 5: Add time offsets to preserve ordering within each content
     _add_temporal_offsets(extracted_facts, contents)
 
-    # Step 5: Auto-tag facts from label groups with tag=True
+    # Step 6: Auto-tag facts from label groups with tag=True
     _inject_label_tags(extracted_facts, config)
 
     return extracted_facts, chunks_metadata, total_usage
 
 
+def _collapse_to_verbatim(facts: list[ExtractedFactType], chunks: list[ChunkMetadata]) -> list[ExtractedFactType]:
+    """
+    For verbatim mode: ensure one fact per chunk with the original chunk text preserved.
+
+    The LLM prompt asks for exactly one fact per chunk, but if it returns more,
+    this collapses them: keeps the first fact as representative, overrides its
+    fact_text with the raw chunk text, and merges entities from any extra facts.
+    """
+    chunk_text_map = {c.chunk_index: c.chunk_text for c in chunks}
+    seen: dict[int, ExtractedFactType] = {}
+    result: list[ExtractedFactType] = []
+
+    for fact in facts:
+        if fact.chunk_index not in seen:
+            fact.fact_text = chunk_text_map.get(fact.chunk_index, fact.fact_text)
+            seen[fact.chunk_index] = fact
+            result.append(fact)
+        else:
+            # Merge entities from extra facts into the representative
+            representative = seen[fact.chunk_index]
+            for entity in fact.entities:
+                if entity not in representative.entities:
+                    representative.entities.append(entity)
+
+    return result
+
+
 def _parse_datetime(date_str: str):
     """Parse ISO datetime string."""
     from dateutil import parser as date_parser

hindsight-api-slim/hindsight_api/engine/retain/fact_storage.py

@@ -10,13 +10,30 @@ import uuid
 
 from ...config import get_config
 from ..memory_engine import fq_table
-from .bank_utils import DEFAULT_DISPOSITION, create_bank_hnsw_indexes
+from .bank_utils import DEFAULT_DISPOSITION, create_bank_vector_indexes
 from .fact_extraction import _sanitize_text
 from .types import ProcessedFact
 
 logger = logging.getLogger(__name__)
 
 
+async def get_document_content(
+    conn,
+    bank_id: str,
+    document_id: str,
+) -> str | None:
+    """Fetch the original_text of an existing document.
+
+    Returns None if the document does not exist.
+    """
+    row = await conn.fetchval(
+        f"SELECT original_text FROM {fq_table('documents')} WHERE id = $1 AND bank_id = $2",
+        document_id,
+        bank_id,
+    )
+    return row
+
+
 async def insert_facts_batch(
     conn, bank_id: str, facts: list[ProcessedFact], document_id: str | None = None
 ) -> list[str]:
@@ -44,7 +61,6 @@ async def insert_facts_batch(
     mentioned_ats = []
     contexts = []
     fact_types = []
-    confidence_scores = []
     metadata_jsons = []
     chunk_ids = []
     document_ids = []
@@ -64,8 +80,6 @@ async def insert_facts_batch(
         mentioned_ats.append(fact.mentioned_at)
         contexts.append(_sanitize_text(fact.context))
         fact_types.append(fact.fact_type)
-        # confidence_score is only for opinion facts
-        confidence_scores.append(1.0 if fact.fact_type == "opinion" else None)
         metadata_jsons.append(json.dumps(fact.metadata))
         chunk_ids.append(fact.chunk_id)
         # Use per-fact document_id if available, otherwise fallback to batch-level document_id
@@ -81,9 +95,15 @@ async def insert_facts_batch(
         if fact.entities:
             signal_parts.extend(e.name for e in fact.entities)
         if fact.occurred_start:
-            signal_parts.append(fact.occurred_start.strftime("%B %-d %Y"))
+            try:
+                signal_parts.append(fact.occurred_start.strftime("%B %d %Y").lstrip("0").replace(" 0", " "))
+            except (ValueError, AttributeError):
+                pass
         if fact.occurred_end and fact.occurred_end != fact.occurred_start:
-            signal_parts.append(fact.occurred_end.strftime("%B %-d %Y"))
+            try:
+                signal_parts.append(fact.occurred_end.strftime("%B %d %Y").lstrip("0").replace(" 0", " "))
+            except (ValueError, AttributeError):
+                pass
         text_signals_list.append(" ".join(signal_parts) if signal_parts else None)
 
     # Batch insert all facts
@@ -97,18 +117,18 @@ async def insert_facts_batch(
             WITH input_data AS (
                 SELECT * FROM unnest(
                     $2::text[], $3::vector[], $4::timestamptz[], $5::timestamptz[], $6::timestamptz[], $7::timestamptz[],
-                    $8::text[], $9::text[], $10::float[], $11::jsonb[], $12::text[], $13::text[], $14::jsonb[], $15::jsonb[], $16::text[]
+                    $8::text[], $9::text[], $10::jsonb[], $11::text[], $12::text[], $13::jsonb[], $14::jsonb[], $15::text[]
                 ) AS t(text, embedding, event_date, occurred_start, occurred_end, mentioned_at,
-                       context, fact_type, confidence_score, metadata, chunk_id, document_id, tags_json,
+                       context, fact_type, metadata, chunk_id, document_id, tags_json,
                        observation_scopes_json, text_signals)
             )
             INSERT INTO {fq_table("memory_units")} (bank_id, text, embedding, event_date, occurred_start, occurred_end, mentioned_at,
-                                     context, fact_type, confidence_score, metadata, chunk_id, document_id, tags,
+                                     context, fact_type, metadata, chunk_id, document_id, tags,
                                      observation_scopes, text_signals, search_vector)
             SELECT
                 $1,
                 text, embedding, event_date, occurred_start, occurred_end, mentioned_at,
-                context, fact_type, confidence_score, metadata, chunk_id, document_id,
+                context, fact_type, metadata, chunk_id, document_id,
                 COALESCE(
                     (SELECT array_agg(elem) FROM jsonb_array_elements_text(tags_json) AS elem),
                     '{{}}'::varchar[]
@@ -129,18 +149,18 @@ async def insert_facts_batch(
             WITH input_data AS (
                 SELECT * FROM unnest(
                     $2::text[], $3::vector[], $4::timestamptz[], $5::timestamptz[], $6::timestamptz[], $7::timestamptz[],
-                    $8::text[], $9::text[], $10::float[], $11::jsonb[], $12::text[], $13::text[], $14::jsonb[], $15::jsonb[], $16::text[]
+                    $8::text[], $9::text[], $10::jsonb[], $11::text[], $12::text[], $13::jsonb[], $14::jsonb[], $15::text[]
                 ) AS t(text, embedding, event_date, occurred_start, occurred_end, mentioned_at,
-                       context, fact_type, confidence_score, metadata, chunk_id, document_id, tags_json,
+                       context, fact_type, metadata, chunk_id, document_id, tags_json,
                        observation_scopes_json, text_signals)
             )
             INSERT INTO {fq_table("memory_units")} (bank_id, text, embedding, event_date, occurred_start, occurred_end, mentioned_at,
-                                     context, fact_type, confidence_score, metadata, chunk_id, document_id, tags,
+                                     context, fact_type, metadata, chunk_id, document_id, tags,
                                      observation_scopes, text_signals)
             SELECT
                 $1,
                 text, embedding, event_date, occurred_start, occurred_end, mentioned_at,
-                context, fact_type, confidence_score, metadata, chunk_id, document_id,
+                context, fact_type, metadata, chunk_id, document_id,
                 COALESCE(
                     (SELECT array_agg(elem) FROM jsonb_array_elements_text(tags_json) AS elem),
                     '{{}}'::varchar[]
@@ -162,7 +182,6 @@ async def insert_facts_batch(
         mentioned_ats,
         contexts,
         fact_types,
-        confidence_scores,
         metadata_jsons,
         chunk_ids,
         document_ids,
@@ -201,8 +220,8 @@ async def ensure_bank_exists(conn, bank_id: str) -> None:
         internal_id,
     )
     if inserted:
-        # Fresh insert — create per-bank HNSW indexes
-        await create_bank_hnsw_indexes(conn, bank_id, str(internal_id))
+        # Fresh insert — create per-bank vector indexes
+        await create_bank_vector_indexes(conn, bank_id, str(internal_id))
 
 
 async def handle_document_tracking(
@@ -215,7 +234,10 @@ async def handle_document_tracking(
     document_tags: list[str] | None = None,
 ) -> None:
     """
-    Handle document tracking in the database.
+    Handle document tracking in the database (full-replace mode).
+
+    Deletes the existing document (cascading to all units and links) on the
+    first batch, then inserts the new document record.
 
     Args:
         conn: Database connection
@@ -232,22 +254,58 @@ async def handle_document_tracking(
     combined_content = _sanitize_text(combined_content) or ""
     content_hash = hashlib.sha256(combined_content.encode()).hexdigest()
 
-    # Always delete old document first if it exists (cascades to units and links)
+    # Delete old document first (cascades to units and links)
     # Only delete on the first batch to avoid deleting data we just inserted
     if is_first_batch:
         await conn.fetchval(
-            f"DELETE FROM {fq_table('documents')} WHERE id = $1 AND bank_id = $2 RETURNING id", document_id, bank_id
+            f"DELETE FROM {fq_table('documents')} WHERE id = $1 AND bank_id = $2 RETURNING id",
+            document_id,
+            bank_id,
         )
 
     # Insert document (or update if exists from concurrent operations)
+    await _upsert_document_row(conn, bank_id, document_id, combined_content, content_hash, retain_params, document_tags)
+
+
+async def upsert_document_metadata(
+    conn,
+    bank_id: str,
+    document_id: str,
+    combined_content: str,
+    retain_params: dict | None = None,
+    document_tags: list[str] | None = None,
+) -> None:
+    """
+    Update document metadata without deleting existing facts/chunks.
+
+    Used by delta retain: the document row is upserted but chunks and
+    memory_units are managed separately at the chunk level.
+    """
+    import hashlib
+
+    combined_content = _sanitize_text(combined_content) or ""
+    content_hash = hashlib.sha256(combined_content.encode()).hexdigest()
+
+    await _upsert_document_row(conn, bank_id, document_id, combined_content, content_hash, retain_params, document_tags)
+
+
+async def _upsert_document_row(
+    conn,
+    bank_id: str,
+    document_id: str,
+    combined_content: str,
+    content_hash: str,
+    retain_params: dict | None = None,
+    document_tags: list[str] | None = None,
+) -> None:
+    """Insert or update a document row."""
     await conn.execute(
         f"""
-        INSERT INTO {fq_table("documents")} (id, bank_id, original_text, content_hash, metadata, retain_params, tags)
-        VALUES ($1, $2, $3, $4, $5, $6, $7)
+        INSERT INTO {fq_table("documents")} (id, bank_id, original_text, content_hash, retain_params, tags)
+        VALUES ($1, $2, $3, $4, $5, $6)
         ON CONFLICT (id, bank_id) DO UPDATE
         SET original_text = EXCLUDED.original_text,
             content_hash = EXCLUDED.content_hash,
-            metadata = EXCLUDED.metadata,
             retain_params = EXCLUDED.retain_params,
             tags = EXCLUDED.tags,
             updated_at = NOW()
@@ -256,7 +314,37 @@ async def handle_document_tracking(
         bank_id,
         combined_content,
         content_hash,
-        json.dumps({}),  # Empty metadata dict
         json.dumps(retain_params) if retain_params else None,
         document_tags or [],
     )
+
+
+async def update_memory_units_tags(
+    conn,
+    bank_id: str,
+    document_id: str,
+    tags: list[str],
+) -> int:
+    """
+    Update tags on all memory_units belonging to a document.
+
+    Used during delta retain to propagate tag changes to unchanged facts.
+
+    Returns:
+        Number of memory units updated.
+    """
+    result = await conn.execute(
+        f"""
+        UPDATE {fq_table("memory_units")}
+        SET tags = $3, updated_at = NOW()
+        WHERE bank_id = $1 AND document_id = $2
+        """,
+        bank_id,
+        document_id,
+        tags or [],
+    )
+    # result is a status string like "UPDATE 5"
+    try:
+        return int(result.split()[-1])
+    except (ValueError, IndexError):
+        return 0

hindsight-api-slim/hindsight_api/engine/retain/link_creation.py

@@ -32,17 +32,26 @@ async def create_temporal_links_batch(conn, bank_id: str, unit_ids: list[str]) -
     return await link_utils.create_temporal_links_batch_per_fact(conn, bank_id, unit_ids, log_buffer=[])
 
 
-async def create_semantic_links_batch(conn, bank_id: str, unit_ids: list[str], embeddings: list[list[float]]) -> int:
+async def create_semantic_links_batch(
+    conn,
+    bank_id: str,
+    unit_ids: list[str],
+    embeddings: list[list[float]],
+    pre_computed_ann_links: list[tuple] | None = None,
+) -> int:
     """
     Create semantic links between facts.
 
     Links facts that are semantically similar based on embeddings.
+    When pre_computed_ann_links are provided (from Phase 1), they are used
+    instead of running ANN queries inside the transaction.
 
     Args:
         conn: Database connection
         bank_id: Bank identifier
         unit_ids: List of unit IDs to create links for
         embeddings: List of embedding vectors (same length as unit_ids)
+        pre_computed_ann_links: Pre-computed ANN results from Phase 1
 
     Returns:
         Number of semantic links created
@@ -53,10 +62,12 @@ async def create_semantic_links_batch(conn, bank_id: str, unit_ids: list[str], e
     if len(unit_ids) != len(embeddings):
         raise ValueError(f"Mismatch between unit_ids ({len(unit_ids)}) and embeddings ({len(embeddings)})")
 
-    return await link_utils.create_semantic_links_batch(conn, bank_id, unit_ids, embeddings, log_buffer=[])
+    return await link_utils.create_semantic_links_batch(
+        conn, bank_id, unit_ids, embeddings, log_buffer=[], pre_computed_ann_links=pre_computed_ann_links
+    )
 
 
-async def create_causal_links_batch(conn, unit_ids: list[str], facts: list[ProcessedFact]) -> int:
+async def create_causal_links_batch(conn, bank_id: str, unit_ids: list[str], facts: list[ProcessedFact]) -> int:
     """
     Create causal links between facts.
 
@@ -94,6 +105,6 @@ async def create_causal_links_batch(conn, unit_ids: list[str], facts: list[Proce
         else:
             causal_relations_per_fact.append([])
 
-    link_count = await link_utils.create_causal_links_batch(conn, unit_ids, causal_relations_per_fact)
+    link_count = await link_utils.create_causal_links_batch(conn, bank_id, unit_ids, causal_relations_per_fact)
 
     return link_count

hindsight-api-slim/hindsight_api/engine/retain/types.py

@@ -25,6 +25,9 @@ class RetainContentDict(TypedDict, total=False):
         observation_scopes: How to scope observations for consolidation (optional).
             "per_tag" runs one pass per individual tag; "combined" (default) runs a
             single pass with all tags; a list[list[str]] specifies exact passes.
+        update_mode: How to handle existing documents with the same document_id (optional).
+            "replace" (default) deletes old data and reprocesses. "append" concatenates
+            new content to the existing document and reprocesses.
     """
 
     content: str  # Required
@@ -37,6 +40,7 @@ class RetainContentDict(TypedDict, total=False):
     observation_scopes: (
         Literal["per_tag", "combined", "all_combinations"] | list[list[str]]
     )  # Observation scopes for consolidation
+    update_mode: Literal["replace", "append"]
 
 
 @dataclass
@@ -107,7 +111,7 @@ class ExtractedFact:
     """
 
     fact_text: str
-    fact_type: str  # "world", "experience", "opinion", "observation"
+    fact_type: str  # "world", "experience", "observation"
     entities: list[str] = field(default_factory=list)
     occurred_start: datetime | None = None
     occurred_end: datetime | None = None
@@ -221,6 +225,45 @@ class ProcessedFact:
         )
 
 
+@dataclass
+class Phase3Context:
+    """
+    Data passed from Phase 2 to Phase 3 for entity link building.
+
+    Contains the unit IDs and entity resolution data needed to build
+    entity links for UI graph visualization after the write transaction commits.
+    """
+
+    unit_ids: list[str] = field(default_factory=list)
+    resolved_entity_ids: list[str] = field(default_factory=list)
+    entity_to_unit: list[tuple] = field(default_factory=list)
+    unit_to_entity_ids: dict[str, list[str]] = field(default_factory=dict)
+
+
+@dataclass
+class EntityResolutionResult:
+    """
+    Result of Phase 1 entity resolution.
+
+    Contains resolved entity IDs and the mapping data needed to remap
+    placeholder unit IDs to real IDs after fact insertion in Phase 2.
+    """
+
+    resolved_entity_ids: list[str]
+    entity_to_unit: list[tuple]
+    unit_to_entity_ids: dict[str, list[str]]
+
+
+@dataclass
+class Phase1Result:
+    """
+    Full result of Phase 1 (entity resolution + optional semantic ANN).
+    """
+
+    entities: EntityResolutionResult
+    semantic_ann_links: list[tuple]
+
+
 @dataclass
 class EntityLink:
     """
@@ -248,7 +291,6 @@ class RetainBatch:
     contents: list[RetainContent]
     document_id: str | None = None
     fact_type_override: str | None = None
-    confidence_score: float | None = None
     document_tags: list[str] = field(default_factory=list)  # Tags applied to all items
 
     # Extracted data (populated during processing)

hindsight-api-slim/hindsight_api/engine/search/__init__.py

@@ -3,12 +3,11 @@ Search module for memory retrieval.
 
 Provides modular search architecture:
 - Retrieval: 4-way parallel (semantic + BM25 + graph + temporal)
-- Graph retrieval: Pluggable strategies (BFS, PPR)
+- Graph retrieval: Link expansion strategy
 - Reranking: Pluggable strategies (heuristic, cross-encoder)
 """
 
-from .graph_retrieval import BFSGraphRetriever, GraphRetriever
-from .mpfp_retrieval import MPFPGraphRetriever
+from .graph_retrieval import GraphRetriever
 from .reranking import CrossEncoderReranker
 from .retrieval import (
     ParallelRetrievalResult,
@@ -21,7 +20,5 @@ __all__ = [
     "set_default_graph_retriever",
     "ParallelRetrievalResult",
     "GraphRetriever",
-    "BFSGraphRetriever",
-    "MPFPGraphRetriever",
     "CrossEncoderReranker",
 ]

hindsight-api-slim/hindsight_api/engine/search/graph_retrieval.py

@@ -2,17 +2,15 @@
 Graph retrieval strategies for memory recall.
 
 This module provides an abstraction for graph-based memory retrieval,
-allowing different algorithms (BFS spreading activation, PPR, etc.) to be
-swapped without changing the rest of the recall pipeline.
+allowing different algorithms to be swapped without changing the rest
+of the recall pipeline.
 """
 
 import logging
 from abc import ABC, abstractmethod
 
-from ..db_utils import acquire_with_retry
-from ..memory_engine import fq_table
-from .tags import TagsMatch, filter_results_by_tags
-from .types import MPFPTimings, RetrievalResult
+from .tags import TagGroup, TagsMatch
+from .types import GraphRetrievalTimings, RetrievalResult
 
 logger = logging.getLogger(__name__)
 
@@ -29,7 +27,7 @@ class GraphRetriever(ABC):
     @property
     @abstractmethod
     def name(self) -> str:
-        """Return identifier for this retrieval strategy (e.g., 'bfs', 'mpfp')."""
+        """Return identifier for this retrieval strategy (e.g., 'link_expansion')."""
         pass
 
     @abstractmethod
@@ -46,7 +44,8 @@ class GraphRetriever(ABC):
         adjacency=None,  # TypedAdjacency, optional pre-loaded graph
         tags: list[str] | None = None,  # Visibility scope tags for filtering
         tags_match: TagsMatch = "any",  # How to match tags: 'any' (OR) or 'all' (AND)
-    ) -> tuple[list[RetrievalResult], MPFPTimings | None]:
+        tag_groups: list[TagGroup] | None = None,  # Compound boolean tag filter groups
+    ) -> tuple[list[RetrievalResult], GraphRetrievalTimings | None]:
         """
         Retrieve relevant facts via graph traversal.
 
@@ -54,211 +53,15 @@ class GraphRetriever(ABC):
             pool: Database connection pool
             query_embedding_str: Query embedding as string (for finding entry points)
             bank_id: Memory bank identifier
-            fact_type: Fact type to filter ('world', 'experience', 'opinion', 'observation')
+            fact_type: Fact type to filter ('world', 'experience', 'observation')
             budget: Maximum number of nodes to explore/return
             query_text: Original query text (optional, for some strategies)
             semantic_seeds: Pre-computed semantic entry points (from semantic retrieval)
             temporal_seeds: Pre-computed temporal entry points (from temporal retrieval)
-            adjacency: Pre-loaded typed adjacency graph (optional, for MPFP)
+            adjacency: Pre-loaded typed adjacency graph (optional)
             tags: Optional list of tags for visibility filtering (OR matching)
 
         Returns:
             Tuple of (List of RetrievalResult with activation scores, optional timing info)
         """
         pass
-
-
-class BFSGraphRetriever(GraphRetriever):
-    """
-    Graph retrieval using BFS-style spreading activation.
-
-    Starting from semantic entry points, spreads activation through
-    the memory graph (entity, temporal, causal links) using breadth-first
-    traversal with decaying activation.
-
-    This is the original Hindsight graph retrieval algorithm.
-    """
-
-    def __init__(
-        self,
-        entry_point_limit: int = 5,
-        entry_point_threshold: float = 0.5,
-        activation_decay: float = 0.8,
-        min_activation: float = 0.1,
-        batch_size: int = 20,
-    ):
-        """
-        Initialize BFS graph retriever.
-
-        Args:
-            entry_point_limit: Maximum number of entry points to start from
-            entry_point_threshold: Minimum semantic similarity for entry points
-            activation_decay: Decay factor per hop (activation *= decay)
-            min_activation: Minimum activation to continue spreading
-            batch_size: Number of nodes to process per batch (for neighbor fetching)
-        """
-        self.entry_point_limit = entry_point_limit
-        self.entry_point_threshold = entry_point_threshold
-        self.activation_decay = activation_decay
-        self.min_activation = min_activation
-        self.batch_size = batch_size
-
-    @property
-    def name(self) -> str:
-        return "bfs"
-
-    async def retrieve(
-        self,
-        pool,
-        query_embedding_str: str,
-        bank_id: str,
-        fact_type: str,
-        budget: int,
-        query_text: str | None = None,
-        semantic_seeds: list[RetrievalResult] | None = None,
-        temporal_seeds: list[RetrievalResult] | None = None,
-        adjacency=None,  # Not used by BFS
-        tags: list[str] | None = None,
-        tags_match: TagsMatch = "any",
-    ) -> tuple[list[RetrievalResult], MPFPTimings | None]:
-        """
-        Retrieve facts using BFS spreading activation.
-
-        Algorithm:
-        1. Find entry points (top semantic matches above threshold)
-        2. BFS traversal: visit neighbors, propagate decaying activation
-        3. Boost causal links (causes, enables, prevents)
-        4. Return visited nodes up to budget
-
-        Note: BFS finds its own entry points via embedding search.
-        The semantic_seeds, temporal_seeds, and adjacency parameters are accepted
-        for interface compatibility but not used.
-        """
-        async with acquire_with_retry(pool) as conn:
-            results = await self._retrieve_with_conn(
-                conn, query_embedding_str, bank_id, fact_type, budget, tags=tags, tags_match=tags_match
-            )
-            return results, None
-
-    async def _retrieve_with_conn(
-        self,
-        conn,
-        query_embedding_str: str,
-        bank_id: str,
-        fact_type: str,
-        budget: int,
-        tags: list[str] | None = None,
-        tags_match: TagsMatch = "any",
-    ) -> list[RetrievalResult]:
-        """Internal implementation with connection."""
-        from .tags import build_tags_where_clause_simple
-
-        tags_clause = build_tags_where_clause_simple(tags, 6, match=tags_match)
-        params = [query_embedding_str, bank_id, fact_type, self.entry_point_threshold, self.entry_point_limit]
-        if tags:
-            params.append(tags)
-
-        # Step 1: Find entry points
-        entry_points = await conn.fetch(
-            f"""
-            SELECT id, text, context, event_date, occurred_start, occurred_end,
-                   mentioned_at, fact_type, document_id, chunk_id, tags,
-                   1 - (embedding <=> $1::vector) AS similarity
-            FROM {fq_table("memory_units")}
-            WHERE bank_id = $2
-              AND embedding IS NOT NULL
-              AND fact_type = $3
-              AND (1 - (embedding <=> $1::vector)) >= $4
-              {tags_clause}
-            ORDER BY embedding <=> $1::vector
-            LIMIT $5
-            """,
-            *params,
-        )
-
-        if not entry_points:
-            logger.debug(
-                f"[BFS] No entry points found for fact_type={fact_type} (tags={tags}, tags_match={tags_match})"
-            )
-            return []
-
-        logger.debug(
-            f"[BFS] Found {len(entry_points)} entry points for fact_type={fact_type} "
-            f"(tags={tags}, tags_match={tags_match})"
-        )
-
-        # Step 2: BFS spreading activation
-        visited = set()
-        results = []
-        queue = [(RetrievalResult.from_db_row(dict(r)), r["similarity"]) for r in entry_points]
-        budget_remaining = budget
-
-        while queue and budget_remaining > 0:
-            # Collect a batch of nodes to process
-            batch_nodes = []
-            batch_activations = {}
-
-            while queue and len(batch_nodes) < self.batch_size and budget_remaining > 0:
-                current, activation = queue.pop(0)
-                unit_id = current.id
-
-                if unit_id not in visited:
-                    visited.add(unit_id)
-                    budget_remaining -= 1
-                    current.activation = activation
-                    results.append(current)
-                    batch_nodes.append(current.id)
-                    batch_activations[unit_id] = activation
-
-            # Batch fetch neighbors
-            if batch_nodes and budget_remaining > 0:
-                max_neighbors = len(batch_nodes) * 20
-                neighbors = await conn.fetch(
-                    f"""
-                    SELECT mu.id, mu.text, mu.context, mu.occurred_start, mu.occurred_end,
-                           mu.mentioned_at, mu.fact_type,
-                           mu.document_id, mu.chunk_id, mu.tags,
-                           ml.weight, ml.link_type, ml.from_unit_id
-                    FROM {fq_table("memory_links")} ml
-                    JOIN {fq_table("memory_units")} mu ON ml.to_unit_id = mu.id
-                    WHERE ml.from_unit_id = ANY($1::uuid[])
-                      AND ml.weight >= $2
-                      AND mu.fact_type = $3
-                    ORDER BY ml.weight DESC
-                    LIMIT $4
-                    """,
-                    batch_nodes,
-                    self.min_activation,
-                    fact_type,
-                    max_neighbors,
-                )
-
-                for n in neighbors:
-                    neighbor_id = str(n["id"])
-                    if neighbor_id not in visited:
-                        parent_id = str(n["from_unit_id"])
-                        parent_activation = batch_activations.get(parent_id, 0.5)
-
-                        # Boost causal links
-                        link_type = n["link_type"]
-                        base_weight = n["weight"]
-
-                        if link_type in ("causes", "caused_by"):
-                            causal_boost = 2.0
-                        elif link_type in ("enables", "prevents"):
-                            causal_boost = 1.5
-                        else:
-                            causal_boost = 1.0
-
-                        effective_weight = base_weight * causal_boost
-                        new_activation = parent_activation * effective_weight * self.activation_decay
-
-                        if new_activation > self.min_activation:
-                            neighbor_result = RetrievalResult.from_db_row(dict(n))
-                            queue.append((neighbor_result, new_activation))
-
-        # Apply tags filtering (BFS may traverse into memories that don't match tags criteria)
-        if tags:
-            results = filter_results_by_tags(results, tags, match=tags_match)
-
-        return results

hindsight-api-slim/hindsight_api/engine/search/link_expansion_retrieval.py

@@ -4,32 +4,37 @@ Link Expansion graph retrieval.
 Expands from semantic/temporal seeds through three parallel, first-class signals
 stored in memory_links:
 
-1. Entity links  — precomputed co-occurrence graph (created at retain time, bounded to
-                   MAX_LINKS_PER_ENTITY per entity). Score = number of distinct shared
-                   entities between the seed set and each candidate.
+1. Entity links  — query-time self-join through unit_entities. Score = number of distinct
+                   shared entities between the seed set and each candidate, computed via
+                   COUNT(DISTINCT entity_id). Uses a LATERAL per-entity cap
+                   (graph_per_entity_limit, default 200) to prevent high-fanout entities
+                   from exploding the self-join intermediate rows.
 2. Semantic links — precomputed kNN graph (each new fact linked to its top-5 most
                     similar existing facts at insert time, similarity >= 0.7). Checked
                     in both directions since the graph is not symmetric. Score = weight.
 3. Causal links  — explicit causal chains (causes/caused_by/enables/prevents).
                    Score = weight + 1.0 (boosted as highest-quality signal).
 
-All three signals are bounded at retain time, so no LATERAL fan-out caps are needed
-at query time. Each expansion is a simple aggregation over a small result set.
+Entity expansion is bounded by graph_per_entity_limit (LATERAL cap per entity).
+A timeout fallback (graph_expansion_timeout) drops entity expansion entirely if the
+query still exceeds the budget.
 
 For non-observation fact types the three expansions are issued as a single CTE query
 (one roundtrip, one connection) with a `source` discriminator column so the Python
 merge step can apply per-signal score transformations.
 """
 
+import asyncio
 import logging
 import math
 import time
 
+from ...config import get_config
 from ..db_utils import acquire_with_retry
 from ..memory_engine import fq_table
 from .graph_retrieval import GraphRetriever
-from .tags import TagsMatch, filter_results_by_tags
-from .types import MPFPTimings, RetrievalResult
+from .tags import TagGroup, TagsMatch, filter_results_by_tag_groups, filter_results_by_tags
+from .types import GraphRetrievalTimings, RetrievalResult
 
 logger = logging.getLogger(__name__)
 
@@ -43,19 +48,23 @@ async def _find_semantic_seeds(
     threshold: float = 0.3,
     tags: list[str] | None = None,
     tags_match: TagsMatch = "any",
+    tag_groups: list[TagGroup] | None = None,
 ) -> list[RetrievalResult]:
     """Find semantic seeds via embedding search."""
-    from .tags import build_tags_where_clause_simple
+    from .tags import build_tag_groups_where_clause, build_tags_where_clause_simple
 
     tags_clause = build_tags_where_clause_simple(tags, 6, match=tags_match)
+    tag_groups_param_start = 6 + (1 if tags else 0)
+    groups_clause, groups_params, _ = build_tag_groups_where_clause(tag_groups, tag_groups_param_start)
     params = [query_embedding_str, bank_id, fact_type, threshold, limit]
     if tags:
         params.append(tags)
+    params.extend(groups_params)
 
     rows = await conn.fetch(
         f"""
         SELECT id, text, context, event_date, occurred_start, occurred_end,
-               mentioned_at, fact_type, document_id, chunk_id, tags,
+               mentioned_at, fact_type, document_id, chunk_id, tags, proof_count,
                1 - (embedding <=> $1::vector) AS similarity
         FROM {fq_table("memory_units")}
         WHERE bank_id = $2
@@ -63,6 +72,7 @@ async def _find_semantic_seeds(
           AND fact_type = $3
           AND (1 - (embedding <=> $1::vector)) >= $4
           {tags_clause}
+          {groups_clause}
         ORDER BY embedding <=> $1::vector
         LIMIT $5
         """,
@@ -110,7 +120,8 @@ class LinkExpansionRetriever(GraphRetriever):
         adjacency=None,
         tags: list[str] | None = None,
         tags_match: TagsMatch = "any",
-    ) -> tuple[list[RetrievalResult], MPFPTimings | None]:
+        tag_groups: list[TagGroup] | None = None,
+    ) -> tuple[list[RetrievalResult], GraphRetrievalTimings | None]:
         """
         Retrieve facts by expanding links from seeds.
 
@@ -130,7 +141,7 @@ class LinkExpansionRetriever(GraphRetriever):
             Tuple of (results, timings)
         """
         start_time = time.time()
-        timings = MPFPTimings(fact_type=fact_type)
+        timings = GraphRetrievalTimings(fact_type=fact_type)
 
         async with acquire_with_retry(pool) as conn:
             # Find seeds if not provided
@@ -147,6 +158,7 @@ class LinkExpansionRetriever(GraphRetriever):
                     threshold=0.3,
                     tags=tags,
                     tags_match=tags_match,
+                    tag_groups=tag_groups,
                 )
                 timings.seeds_time = time.time() - seeds_start
                 logger.debug(
@@ -221,6 +233,9 @@ class LinkExpansionRetriever(GraphRetriever):
         if tags:
             results = filter_results_by_tags(results, tags, match=tags_match)
 
+        if tag_groups:
+            results = filter_results_by_tag_groups(results, tag_groups)
+
         timings.result_count = len(results)
         timings.traverse = time.time() - start_time
 
@@ -252,31 +267,48 @@ class LinkExpansionRetriever(GraphRetriever):
                     idx_memory_links_to_type_weight (to_unit_id, link_type, weight DESC)
                     → replaces costly BitmapAnd of two separate scans
         """
+        config = get_config()
         ml = fq_table("memory_links")
         mu = fq_table("memory_units")
-        all_rows = await conn.fetch(
-            f"""
-            WITH entity_expanded AS (
-                -- Entity co-occurrence: seeds → their precomputed entity-link neighbors.
-                -- Score = distinct shared entities (bounded at retain time to
-                -- MAX_LINKS_PER_ENTITY=50).  GROUP BY mu.id is sufficient because mu.id
-                -- is the primary key and functionally determines all other mu columns.
-                SELECT
-                    mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
-                    mu.occurred_end, mu.mentioned_at,
-                    mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
-                    COUNT(DISTINCT ml.entity_id)::float AS score,
-                    'entity'::text AS source
-                FROM {ml} ml
-                JOIN {mu} mu ON mu.id = ml.to_unit_id
-                WHERE ml.from_unit_id = ANY($1::uuid[])
-                  AND ml.link_type = 'entity'
-                  AND mu.fact_type = $2
-                  AND mu.id != ALL($1::uuid[])
+        ue = fq_table("unit_entities")
+
+        per_entity_limit = config.link_expansion_per_entity_limit
+
+        # Entity CTE with LATERAL fanout cap.
+        # Every seed entity (including high-frequency ones) is kept, but each
+        # entity's expansion is capped to per_entity_limit target units.  The
+        # LATERAL subquery orders by unit_id DESC so the most recently inserted
+        # units are preferred (a recency proxy that is free — it rides the PK
+        # index with no extra sort).
+        entity_cte = f"""
+            seed_entities AS (
+                SELECT DISTINCT ue.entity_id
+                FROM {ue} ue
+                WHERE ue.unit_id = ANY($1::uuid[])
+            ),
+            entity_expanded AS (
+                SELECT mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
+                       mu.occurred_end, mu.mentioned_at,
+                       mu.fact_type, mu.document_id, mu.chunk_id, mu.tags, mu.proof_count,
+                       COUNT(DISTINCT se.entity_id)::float AS score,
+                       'entity'::text AS source
+                FROM seed_entities se
+                CROSS JOIN LATERAL (
+                    SELECT ue_target.unit_id
+                    FROM {ue} ue_target
+                    WHERE ue_target.entity_id = se.entity_id
+                      AND ue_target.unit_id != ALL($1::uuid[])
+                    ORDER BY ue_target.unit_id DESC
+                    LIMIT {per_entity_limit}
+                ) t
+                JOIN {mu} mu ON mu.id = t.unit_id
+                WHERE mu.fact_type = $2
                 GROUP BY mu.id
                 ORDER BY score DESC
                 LIMIT $3
-            ),
+            )"""
+
+        semantic_causal_cte = f"""
             semantic_expanded AS (
                 -- Semantic kNN: both outgoing (seeds → their kNN at insert time) and
                 -- incoming (facts inserted after seeds that found seeds as kNN).
@@ -284,14 +316,14 @@ class LinkExpansionRetriever(GraphRetriever):
                 SELECT
                     id, text, context, event_date, occurred_start,
                     occurred_end, mentioned_at,
-                    fact_type, document_id, chunk_id, tags,
+                    fact_type, document_id, chunk_id, tags, proof_count,
                     MAX(weight) AS score,
                     'semantic'::text AS source
                 FROM (
                     SELECT
                         mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                         mu.occurred_end, mu.mentioned_at,
-                        mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
+                        mu.fact_type, mu.document_id, mu.chunk_id, mu.tags, mu.proof_count,
                         ml.weight
                     FROM {ml} ml
                     JOIN {mu} mu ON mu.id = ml.to_unit_id
@@ -303,7 +335,7 @@ class LinkExpansionRetriever(GraphRetriever):
                     SELECT
                         mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                         mu.occurred_end, mu.mentioned_at,
-                        mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
+                        mu.fact_type, mu.document_id, mu.chunk_id, mu.tags, mu.proof_count,
                         ml.weight
                     FROM {ml} ml
                     JOIN {mu} mu ON mu.id = ml.from_unit_id
@@ -314,7 +346,7 @@ class LinkExpansionRetriever(GraphRetriever):
                 ) sem_raw
                 GROUP BY id, text, context, event_date, occurred_start,
                          occurred_end, mentioned_at,
-                         fact_type, document_id, chunk_id, tags
+                         fact_type, document_id, chunk_id, tags, proof_count
                 ORDER BY score DESC
                 LIMIT $3
             ),
@@ -325,7 +357,7 @@ class LinkExpansionRetriever(GraphRetriever):
                 SELECT DISTINCT ON (mu.id)
                     mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                     mu.occurred_end, mu.mentioned_at,
-                    mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
+                    mu.fact_type, mu.document_id, mu.chunk_id, mu.tags, mu.proof_count,
                     ml.weight AS score,
                     'causal'::text AS source
                 FROM {ml} ml
@@ -336,18 +368,37 @@ class LinkExpansionRetriever(GraphRetriever):
                   AND mu.fact_type = $2
                 ORDER BY mu.id, ml.weight DESC
                 LIMIT $3
-            )
+            )"""
+
+        full_query = f"""
+            WITH {entity_cte},
+            {semantic_causal_cte}
             SELECT * FROM entity_expanded
             UNION ALL
             SELECT * FROM semantic_expanded
             UNION ALL
             SELECT * FROM causal_expanded
-            """,
-            seed_ids,
-            fact_type,
-            budget,
-            self.causal_weight_threshold,
-        )
+            """
+
+        params = [seed_ids, fact_type, budget, self.causal_weight_threshold]
+
+        try:
+            all_rows = await asyncio.wait_for(
+                conn.fetch(full_query, *params),
+                timeout=config.link_expansion_timeout,
+            )
+        except asyncio.TimeoutError:
+            logger.warning(
+                f"[LinkExpansion] Entity expansion timed out after {config.link_expansion_timeout}s "
+                f"for fact_type={fact_type}, falling back to semantic+causal only"
+            )
+            fallback_query = f"""
+                WITH {semantic_causal_cte}
+                SELECT * FROM semantic_expanded
+                UNION ALL
+                SELECT * FROM causal_expanded
+                """
+            all_rows = await conn.fetch(fallback_query, *params)
 
         entity_rows = [r for r in all_rows if r["source"] == "entity"]
         semantic_rows = [r for r in all_rows if r["source"] == "semantic"]
@@ -387,6 +438,33 @@ class LinkExpansionRetriever(GraphRetriever):
                 f"{len(source_ids_found)} source_memory_ids found"
             )
 
+        config = get_config()
+        ue = fq_table("unit_entities")
+        per_entity_limit = config.link_expansion_per_entity_limit
+
+        connected_sources_cte = f"""
+            source_entities AS (
+                SELECT DISTINCT ue_seed.entity_id
+                FROM seed_sources ss
+                JOIN {ue} ue_seed ON ue_seed.unit_id = ss.source_id
+            ),
+            connected_sources AS (
+                -- Find sources sharing entities with seed observation sources
+                -- via LATERAL-capped self-join (prevents hub entity fanout).
+                SELECT DISTINCT t.unit_id AS source_id
+                FROM source_entities se
+                CROSS JOIN LATERAL (
+                    SELECT ue_target.unit_id
+                    FROM {ue} ue_target
+                    WHERE ue_target.entity_id = se.entity_id
+                    ORDER BY ue_target.unit_id DESC
+                    LIMIT {per_entity_limit}
+                ) t
+                WHERE NOT EXISTS (
+                    SELECT 1 FROM seed_sources ss WHERE ss.source_id = t.unit_id
+                )
+            )"""
+
         entity_rows = await conn.fetch(
             f"""
             WITH seed_sources AS (
@@ -395,22 +473,14 @@ class LinkExpansionRetriever(GraphRetriever):
                 WHERE id = ANY($1::uuid[])
                   AND source_memory_ids IS NOT NULL
             ),
-            connected_sources AS (
-                -- Mirror the non-observation entity expansion: follow pre-bounded entity
-                -- links in memory_links (capped to MAX_LINKS_PER_ENTITY=50 at retain time).
-                -- Score = number of distinct shared entities, same as the non-obs path.
-                SELECT DISTINCT ml.to_unit_id AS source_id
-                FROM seed_sources ss
-                JOIN {fq_table("memory_links")} ml ON ml.from_unit_id = ss.source_id
-                WHERE ml.link_type = 'entity'
-            ),
+            {connected_sources_cte},
             connected_array AS (
                 SELECT array_agg(source_id) AS source_ids FROM connected_sources
             )
             SELECT
                 mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                 mu.occurred_end, mu.mentioned_at,
-                mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
+                mu.fact_type, mu.document_id, mu.chunk_id, mu.tags, mu.proof_count,
                 (SELECT COUNT(DISTINCT s) FROM unnest(mu.source_memory_ids) s WHERE s = ANY(ca.source_ids))::float AS score
             FROM {fq_table("memory_units")} mu, connected_array ca
             WHERE mu.fact_type = 'observation'
@@ -434,13 +504,13 @@ class LinkExpansionRetriever(GraphRetriever):
                 SELECT
                     id, text, context, event_date, occurred_start,
                     occurred_end, mentioned_at,
-                    fact_type, document_id, chunk_id, tags,
+                    fact_type, document_id, chunk_id, tags, proof_count,
                     MAX(weight) AS score,
                     'semantic'::text AS source
                 FROM (
                     SELECT mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                            mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.document_id,
-                           mu.chunk_id, mu.tags, ml.weight
+                           mu.chunk_id, mu.tags, mu.proof_count, ml.weight
                     FROM {ml} ml JOIN {mu} mu ON mu.id = ml.to_unit_id
                     WHERE ml.from_unit_id = ANY($1::uuid[])
                       AND ml.link_type = 'semantic' AND mu.fact_type = 'observation'
@@ -448,21 +518,21 @@ class LinkExpansionRetriever(GraphRetriever):
                     UNION ALL
                     SELECT mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                            mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.document_id,
-                           mu.chunk_id, mu.tags, ml.weight
+                           mu.chunk_id, mu.tags, mu.proof_count, ml.weight
                     FROM {ml} ml JOIN {mu} mu ON mu.id = ml.from_unit_id
                     WHERE ml.to_unit_id = ANY($1::uuid[])
                       AND ml.link_type = 'semantic' AND mu.fact_type = 'observation'
                       AND mu.id != ALL($1::uuid[])
                 ) sem_raw
                 GROUP BY id, text, context, event_date, occurred_start, occurred_end,
-                         mentioned_at, fact_type, document_id, chunk_id, tags
+                         mentioned_at, fact_type, document_id, chunk_id, tags, proof_count
                 ORDER BY score DESC LIMIT $2
             ),
             causal_expanded AS (
                 SELECT DISTINCT ON (mu.id)
                     mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start,
                     mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.document_id,
-                    mu.chunk_id, mu.tags, ml.weight AS score, 'causal'::text AS source
+                    mu.chunk_id, mu.tags, mu.proof_count, ml.weight AS score, 'causal'::text AS source
                 FROM {ml} ml JOIN {mu} mu ON ml.to_unit_id = mu.id
                 WHERE ml.from_unit_id = ANY($1::uuid[])
                   AND ml.link_type IN ('causes', 'caused_by', 'enables', 'prevents')

hindsight-api-slim/hindsight_api/engine/search/mpfp_retrieval.py

@@ -1,684 +0,0 @@
-"""
-Meta-Path Forward Push (MPFP) graph retrieval.
-
-A sublinear graph traversal algorithm for memory retrieval over heterogeneous
-graphs with multiple edge types (semantic, temporal, causal, entity).
-
-Combines meta-path patterns from HIN literature with Forward Push local
-propagation from Approximate PPR.
-
-Key properties:
-- Sublinear in graph size (threshold pruning bounds active nodes)
-- Lazy edge loading: only loads edges for frontier nodes, not entire graph
-- Predefined patterns capture different retrieval intents
-- All patterns run in parallel, results fused via RRF
-- No LLM in the loop during traversal
-"""
-
-import asyncio
-import logging
-from collections import defaultdict
-from dataclasses import dataclass, field
-
-from ..db_utils import acquire_with_retry
-from ..memory_engine import fq_table
-from .graph_retrieval import GraphRetriever
-from .tags import TagsMatch
-from .types import MPFPTimings, RetrievalResult
-
-logger = logging.getLogger(__name__)
-
-
-# -----------------------------------------------------------------------------
-# Data Classes
-# -----------------------------------------------------------------------------
-
-
-@dataclass
-class EdgeTarget:
-    """A neighbor node with its edge weight."""
-
-    node_id: str
-    weight: float
-
-
-@dataclass
-class EdgeCache:
-    """
-    Cache for lazily-loaded edges.
-
-    Grows per-hop as edges are loaded for frontier nodes.
-    Shared across patterns to avoid redundant loads.
-    Loads ALL edge types at once to minimize DB queries.
-    Thread-safe via asyncio lock to prevent redundant concurrent loads.
-    """
-
-    # edge_type -> from_node_id -> list of EdgeTarget
-    graphs: dict[str, dict[str, list[EdgeTarget]]] = field(default_factory=dict)
-    # Track which nodes have been fully loaded (all edge types)
-    _fully_loaded: set[str] = field(default_factory=set)
-    # Timing stats
-    db_queries: int = 0
-    edge_load_time: float = 0.0
-    # Detailed hop timing for debugging
-    hop_details: list[dict] = field(default_factory=list)
-    # Lock to prevent redundant concurrent loads
-    _lock: asyncio.Lock = field(default_factory=asyncio.Lock)
-
-    def get_neighbors(self, edge_type: str, node_id: str) -> list[EdgeTarget]:
-        """Get neighbors for a node via a specific edge type."""
-        return self.graphs.get(edge_type, {}).get(node_id, [])
-
-    def get_normalized_neighbors(self, edge_type: str, node_id: str, top_k: int) -> list[EdgeTarget]:
-        """Get top-k neighbors with weights normalized to sum to 1."""
-        neighbors = self.get_neighbors(edge_type, node_id)[:top_k]
-        if not neighbors:
-            return []
-
-        total = sum(n.weight for n in neighbors)
-        if total == 0:
-            return []
-
-        return [EdgeTarget(node_id=n.node_id, weight=n.weight / total) for n in neighbors]
-
-    def is_fully_loaded(self, node_id: str) -> bool:
-        """Check if all edges for this node have been loaded."""
-        return node_id in self._fully_loaded
-
-    def get_uncached(self, node_ids: list[str]) -> list[str]:
-        """Get node IDs that haven't been fully loaded yet."""
-        return [n for n in node_ids if not self.is_fully_loaded(n)]
-
-    def add_all_edges(self, edges_by_type: dict[str, dict[str, list[EdgeTarget]]], all_queried: list[str]):
-        """
-        Add loaded edges to the cache (all edge types at once).
-
-        Args:
-            edges_by_type: Dict mapping edge_type -> from_node_id -> list of EdgeTarget
-            all_queried: All node IDs that were queried (marks them as fully loaded)
-        """
-        for edge_type, edges in edges_by_type.items():
-            if edge_type not in self.graphs:
-                self.graphs[edge_type] = {}
-            for node_id, neighbors in edges.items():
-                self.graphs[edge_type][node_id] = neighbors
-
-        # Mark all queried nodes as fully loaded (even if they have no edges)
-        self._fully_loaded.update(all_queried)
-
-
-@dataclass
-class PatternResult:
-    """Result from a single pattern traversal."""
-
-    pattern: list[str]
-    scores: dict[str, float]  # node_id -> accumulated mass
-
-
-@dataclass
-class MPFPConfig:
-    """Configuration for MPFP algorithm."""
-
-    alpha: float = 0.15  # teleport/keep probability
-    threshold: float = 1e-6  # mass pruning threshold (lower = explore more)
-    top_k_neighbors: int = 20  # fan-out limit per node
-
-    # Patterns from semantic seeds
-    patterns_semantic: list[list[str]] = field(
-        default_factory=lambda: [
-            ["semantic", "semantic"],  # topic expansion
-            ["entity", "temporal"],  # entity timeline
-            ["semantic", "causes"],  # reasoning chains (forward)
-            ["semantic", "caused_by"],  # reasoning chains (backward)
-            ["entity", "semantic"],  # entity context
-        ]
-    )
-
-    # Patterns from temporal seeds
-    patterns_temporal: list[list[str]] = field(
-        default_factory=lambda: [
-            ["temporal", "semantic"],  # what was happening then
-            ["temporal", "entity"],  # who was involved then
-        ]
-    )
-
-
-@dataclass
-class SeedNode:
-    """An entry point node with its initial score."""
-
-    node_id: str
-    score: float  # initial mass (e.g., similarity score)
-
-
-# -----------------------------------------------------------------------------
-# Lazy Edge Loading
-# -----------------------------------------------------------------------------
-
-
-async def load_all_edges_for_frontier(
-    pool,
-    node_ids: list[str],
-    top_k_per_type: int = 20,
-) -> dict[str, dict[str, list[EdgeTarget]]]:
-    """
-    Load top-k edges per (node, edge_type) for frontier nodes.
-
-    Uses a LATERAL join to efficiently fetch only the top-k edges per type,
-    avoiding loading hundreds of entity edges when only 20 are needed.
-
-    Requires composite index: (from_unit_id, link_type, weight DESC)
-
-    Args:
-        pool: Database connection pool
-        node_ids: Frontier node IDs to load edges for
-        top_k_per_type: Max edges to load per (node, link_type) pair
-
-    Returns:
-        Dict mapping edge_type -> from_node_id -> list of EdgeTarget
-    """
-    if not node_ids:
-        return {}
-
-    async with acquire_with_retry(pool) as conn:
-        # Use LATERAL join to get top-k per (from_node, link_type)
-        # This leverages the composite index for efficient early termination
-        rows = await conn.fetch(
-            f"""
-            WITH frontier(node_id) AS (SELECT unnest($1::uuid[]))
-            SELECT f.node_id as from_unit_id, lt.link_type, edges.to_unit_id, edges.weight
-            FROM frontier f
-            CROSS JOIN (VALUES ('semantic'), ('temporal'), ('entity'), ('causes'), ('caused_by')) AS lt(link_type)
-            CROSS JOIN LATERAL (
-                SELECT ml.to_unit_id, ml.weight
-                FROM {fq_table("memory_links")} ml
-                WHERE ml.from_unit_id = f.node_id
-                  AND ml.link_type = lt.link_type
-                  AND ml.weight >= 0.1
-                ORDER BY ml.weight DESC
-                LIMIT $2
-            ) edges
-            """,
-            node_ids,
-            top_k_per_type,
-        )
-
-    # Group by edge_type -> from_node -> neighbors
-    result: dict[str, dict[str, list[EdgeTarget]]] = defaultdict(lambda: defaultdict(list))
-    for row in rows:
-        edge_type = row["link_type"]
-        from_id = str(row["from_unit_id"])
-        to_id = str(row["to_unit_id"])
-        weight = row["weight"]
-        result[edge_type][from_id].append(EdgeTarget(node_id=to_id, weight=weight))
-
-    # Convert nested defaultdicts to regular dicts
-    return {edge_type: dict(edges) for edge_type, edges in result.items()}
-
-
-# -----------------------------------------------------------------------------
-# Core Algorithm (Async with Lazy Loading)
-# -----------------------------------------------------------------------------
-
-
-@dataclass
-class PatternState:
-    """State for a pattern traversal between hops."""
-
-    pattern: list[str]
-    hop_index: int
-    scores: dict[str, float]
-    frontier: dict[str, float]
-
-
-def _init_pattern_state(seeds: list[SeedNode], pattern: list[str]) -> PatternState:
-    """Initialize pattern state from seeds."""
-    if not seeds:
-        return PatternState(pattern=pattern, hop_index=0, scores={}, frontier={})
-
-    total_seed_score = sum(s.score for s in seeds)
-    if total_seed_score == 0:
-        total_seed_score = len(seeds)
-
-    frontier = {s.node_id: s.score / total_seed_score for s in seeds}
-    return PatternState(pattern=pattern, hop_index=0, scores={}, frontier=frontier)
-
-
-def _execute_hop(state: PatternState, cache: EdgeCache, config: MPFPConfig) -> set[str]:
-    """
-    Execute ONE hop of traversal, return frontier nodes for next hop.
-
-    This is a pure function that uses cached edges (no DB access).
-    Returns set of uncached nodes needed for next hop.
-    """
-    if state.hop_index >= len(state.pattern):
-        return set()
-
-    edge_type = state.pattern[state.hop_index]
-
-    # Collect active nodes above threshold
-    active_nodes = [node_id for node_id, mass in state.frontier.items() if mass >= config.threshold]
-    if not active_nodes:
-        state.frontier = {}
-        return set()
-
-    # Propagate mass using cached edges
-    next_frontier: dict[str, float] = {}
-    uncached_for_next: set[str] = set()
-
-    for node_id, mass in state.frontier.items():
-        if mass < config.threshold:
-            continue
-
-        # Keep α portion for this node
-        state.scores[node_id] = state.scores.get(node_id, 0) + config.alpha * mass
-
-        # Push (1-α) to neighbors
-        push_mass = (1 - config.alpha) * mass
-        neighbors = cache.get_normalized_neighbors(edge_type, node_id, config.top_k_neighbors)
-
-        for neighbor in neighbors:
-            next_frontier[neighbor.node_id] = next_frontier.get(neighbor.node_id, 0) + push_mass * neighbor.weight
-            # Track if we'll need edges for this node in the next hop
-            if not cache.is_fully_loaded(neighbor.node_id):
-                uncached_for_next.add(neighbor.node_id)
-
-    state.frontier = next_frontier
-    state.hop_index += 1
-
-    return uncached_for_next
-
-
-def _finalize_pattern(state: PatternState, config: MPFPConfig) -> PatternResult:
-    """Finalize pattern by adding remaining frontier mass to scores."""
-    for node_id, mass in state.frontier.items():
-        if mass >= config.threshold:
-            state.scores[node_id] = state.scores.get(node_id, 0) + mass
-
-    return PatternResult(pattern=state.pattern, scores=state.scores)
-
-
-async def mpfp_traverse_hop_synchronized(
-    pool,
-    pattern_jobs: list[tuple[list[SeedNode], list[str]]],
-    config: MPFPConfig,
-    cache: EdgeCache,
-) -> list[PatternResult]:
-    """
-    Execute ALL patterns with hop-synchronized edge loading.
-
-    Instead of running each pattern independently (causing multiple DB queries),
-    this function:
-    1. Runs hop 1 for ALL patterns (using pre-warmed seed edges)
-    2. Collects ALL unique hop-2 frontier nodes across patterns
-    3. Pre-warms hop-2 edges in ONE query
-    4. Runs hop 2 for ALL patterns
-
-    This reduces DB queries from O(patterns * hops) to O(hops).
-
-    Args:
-        pool: Database connection pool
-        pattern_jobs: List of (seeds, pattern) tuples
-        config: Algorithm parameters
-        cache: Shared edge cache (should be pre-warmed with seed edges)
-
-    Returns:
-        List of PatternResult for each pattern
-    """
-    import time
-
-    # Initialize all pattern states
-    states = [_init_pattern_state(seeds, pattern) for seeds, pattern in pattern_jobs]
-
-    # Determine max hops (all patterns should be same length, but be safe)
-    max_hops = max((len(p) for _, p in pattern_jobs), default=0)
-
-    # Detailed timing for debugging
-    hop_times: list[dict] = []
-
-    # Execute hop-by-hop across ALL patterns
-    for hop in range(max_hops):
-        hop_start = time.time()
-        hop_timing = {"hop": hop, "patterns_executed": 0, "uncached_count": 0, "load_time": 0.0}
-
-        # Execute this hop for all patterns, collect uncached nodes for next hop
-        all_uncached: set[str] = set()
-        exec_start = time.time()
-        for state in states:
-            if state.hop_index < len(state.pattern):
-                uncached = _execute_hop(state, cache, config)
-                all_uncached.update(uncached)
-                hop_timing["patterns_executed"] += 1
-        hop_timing["exec_time"] = time.time() - exec_start
-
-        # Pre-warm edges for ALL uncached nodes before next hop
-        hop_timing["uncached_count"] = len(all_uncached)
-        if all_uncached:
-            uncached_list = list(all_uncached - cache._fully_loaded)
-            hop_timing["uncached_after_filter"] = len(uncached_list)
-            if uncached_list:
-                load_start = time.time()
-                edges_by_type = await load_all_edges_for_frontier(pool, uncached_list, config.top_k_neighbors)
-                hop_timing["load_time"] = time.time() - load_start
-                cache.edge_load_time += hop_timing["load_time"]
-                cache.db_queries += 1
-                cache.add_all_edges(edges_by_type, uncached_list)
-                hop_timing["edges_loaded"] = sum(
-                    len(neighbors) for edges in edges_by_type.values() for neighbors in edges.values()
-                )
-
-        hop_timing["total_time"] = time.time() - hop_start
-        hop_times.append(hop_timing)
-
-    # Store hop timing details in cache for logging
-    cache.hop_details = hop_times
-
-    # Finalize all patterns
-    return [_finalize_pattern(state, config) for state in states]
-
-
-async def mpfp_traverse_async(
-    pool,
-    seeds: list[SeedNode],
-    pattern: list[str],
-    config: MPFPConfig,
-    cache: EdgeCache,
-) -> PatternResult:
-    """
-    Async Forward Push traversal with lazy edge loading.
-
-    NOTE: For better performance with multiple patterns, use mpfp_traverse_hop_synchronized().
-    This function is kept for single-pattern use cases.
-    """
-    if not seeds:
-        return PatternResult(pattern=pattern, scores={})
-
-    results = await mpfp_traverse_hop_synchronized(pool, [(seeds, pattern)], config, cache)
-    return results[0] if results else PatternResult(pattern=pattern, scores={})
-
-
-def rrf_fusion(
-    results: list[PatternResult],
-    k: int = 60,
-    top_k: int = 50,
-) -> list[tuple[str, float]]:
-    """
-    Reciprocal Rank Fusion to combine pattern results.
-
-    Args:
-        results: List of pattern results
-        k: RRF constant (higher = more uniform weighting)
-        top_k: Number of results to return
-
-    Returns:
-        List of (node_id, fused_score) tuples, sorted by score descending
-    """
-    fused: dict[str, float] = {}
-
-    for result in results:
-        if not result.scores:
-            continue
-
-        # Rank nodes by their score in this pattern
-        ranked = sorted(result.scores.keys(), key=lambda n: result.scores[n], reverse=True)
-
-        for rank, node_id in enumerate(ranked):
-            fused[node_id] = fused.get(node_id, 0) + 1.0 / (k + rank + 1)
-
-    # Sort by fused score and return top-k
-    sorted_results = sorted(fused.items(), key=lambda x: x[1], reverse=True)
-
-    return sorted_results[:top_k]
-
-
-# -----------------------------------------------------------------------------
-# Database Loading
-# -----------------------------------------------------------------------------
-
-
-async def fetch_memory_units_by_ids(
-    pool,
-    node_ids: list[str],
-    fact_type: str,
-) -> list[RetrievalResult]:
-    """Fetch full memory unit details for a list of node IDs."""
-    if not node_ids:
-        return []
-
-    async with acquire_with_retry(pool) as conn:
-        rows = await conn.fetch(
-            f"""
-            SELECT id, text, context, event_date, occurred_start, occurred_end,
-                   mentioned_at, fact_type, document_id, chunk_id, tags
-            FROM {fq_table("memory_units")}
-            WHERE id = ANY($1::uuid[])
-              AND fact_type = $2
-            """,
-            node_ids,
-            fact_type,
-        )
-
-    return [RetrievalResult.from_db_row(dict(r)) for r in rows]
-
-
-# -----------------------------------------------------------------------------
-# Graph Retriever Implementation
-# -----------------------------------------------------------------------------
-
-
-class MPFPGraphRetriever(GraphRetriever):
-    """
-    Graph retrieval using Meta-Path Forward Push with lazy edge loading.
-
-    Runs predefined patterns in parallel from semantic and temporal seeds,
-    loading edges on-demand per hop instead of loading entire graph upfront.
-    """
-
-    def __init__(self, config: MPFPConfig | None = None):
-        """
-        Initialize MPFP retriever.
-
-        Args:
-            config: Algorithm configuration (uses defaults if None)
-        """
-        if config is None:
-            # Read top_k_neighbors from global config
-            from ...config import get_config
-
-            global_config = get_config()
-            config = MPFPConfig(top_k_neighbors=global_config.mpfp_top_k_neighbors)
-        self.config = config
-
-    @property
-    def name(self) -> str:
-        return "mpfp"
-
-    async def retrieve(
-        self,
-        pool,
-        query_embedding_str: str,
-        bank_id: str,
-        fact_type: str,
-        budget: int,
-        query_text: str | None = None,
-        semantic_seeds: list[RetrievalResult] | None = None,
-        temporal_seeds: list[RetrievalResult] | None = None,
-        adjacency=None,  # Ignored - kept for interface compatibility
-        tags: list[str] | None = None,
-        tags_match: TagsMatch = "any",
-    ) -> tuple[list[RetrievalResult], MPFPTimings | None]:
-        """
-        Retrieve facts using MPFP algorithm with lazy edge loading.
-
-        Args:
-            pool: Database connection pool
-            query_embedding_str: Query embedding (used for fallback seed finding)
-            bank_id: Memory bank ID
-            fact_type: Fact type to filter
-            budget: Maximum results to return
-            query_text: Original query text (optional)
-            semantic_seeds: Pre-computed semantic entry points
-            temporal_seeds: Pre-computed temporal entry points
-            adjacency: Ignored (kept for interface compatibility)
-            tags: Optional list of tags for visibility filtering (OR matching)
-
-        Returns:
-            Tuple of (List of RetrievalResult with activation scores, MPFPTimings)
-        """
-        import time
-
-        timings = MPFPTimings(fact_type=fact_type)
-
-        # Convert seeds to SeedNode format
-        semantic_seed_nodes = self._convert_seeds(semantic_seeds, "similarity")
-        temporal_seed_nodes = self._convert_seeds(temporal_seeds, "temporal_score")
-
-        # If no semantic seeds provided, fall back to finding our own
-        if not semantic_seed_nodes:
-            seeds_start = time.time()
-            semantic_seed_nodes = await self._find_semantic_seeds(
-                pool, query_embedding_str, bank_id, fact_type, tags=tags, tags_match=tags_match
-            )
-            timings.seeds_time = time.time() - seeds_start
-            logger.debug(
-                f"[MPFP] Found {len(semantic_seed_nodes)} semantic seeds for fact_type={fact_type} (tags={tags}, tags_match={tags_match})"
-            )
-
-        # Collect all pattern jobs
-        pattern_jobs = []
-
-        # Patterns from semantic seeds
-        for pattern in self.config.patterns_semantic:
-            if semantic_seed_nodes:
-                pattern_jobs.append((semantic_seed_nodes, pattern))
-
-        # Patterns from temporal seeds
-        for pattern in self.config.patterns_temporal:
-            if temporal_seed_nodes:
-                pattern_jobs.append((temporal_seed_nodes, pattern))
-
-        if not pattern_jobs:
-            logger.debug(
-                f"[MPFP] No pattern jobs (semantic_seeds={len(semantic_seed_nodes)}, temporal_seeds={len(temporal_seed_nodes)})"
-            )
-            return [], timings
-
-        timings.pattern_count = len(pattern_jobs)
-
-        # Shared edge cache across all patterns
-        cache = EdgeCache()
-
-        # Pre-warm cache with ALL seed node edges BEFORE running patterns
-        # This prevents redundant DB queries at hop 1
-        all_seed_ids = list({s.node_id for seeds, _ in pattern_jobs for s in seeds})
-        if all_seed_ids:
-            import time as time_module
-
-            prewarm_start = time_module.time()
-            edges_by_type = await load_all_edges_for_frontier(pool, all_seed_ids, self.config.top_k_neighbors)
-            cache.edge_load_time += time_module.time() - prewarm_start
-            cache.db_queries += 1
-            cache.add_all_edges(edges_by_type, all_seed_ids)
-
-        # Run all patterns with HOP-SYNCHRONIZED edge loading
-        # This batches hop-2 edge loads across ALL patterns into ONE query
-        # Reduces DB queries from O(patterns * hops) to O(hops)
-        step_start = time.time()
-        pattern_results = await mpfp_traverse_hop_synchronized(pool, pattern_jobs, self.config, cache)
-        timings.traverse = time.time() - step_start
-
-        # Record edge loading stats from cache
-        timings.edge_count = sum(len(neighbors) for g in cache.graphs.values() for neighbors in g.values())
-        timings.db_queries = cache.db_queries
-        timings.edge_load_time = cache.edge_load_time
-        timings.hop_details = cache.hop_details
-
-        # Fuse results
-        step_start = time.time()
-        fused = rrf_fusion(pattern_results, top_k=budget)
-        timings.fusion = time.time() - step_start
-
-        if not fused:
-            logger.debug(f"[MPFP] No fused results after RRF fusion (pattern_count={len(pattern_results)})")
-            return [], timings
-
-        # Get top result IDs
-        result_ids = [node_id for node_id, score in fused][:budget]
-
-        # Fetch full details
-        step_start = time.time()
-        results = await fetch_memory_units_by_ids(pool, result_ids, fact_type)
-        timings.fetch = time.time() - step_start
-
-        # Filter results by tags (graph traversal may have picked up unfiltered memories)
-        if tags:
-            from .tags import filter_results_by_tags
-
-            results = filter_results_by_tags(results, tags, match=tags_match)
-
-        timings.result_count = len(results)
-
-        # Add activation scores from fusion
-        score_map = {node_id: score for node_id, score in fused}
-        for result in results:
-            result.activation = score_map.get(result.id, 0.0)
-
-        # Sort by activation
-        results.sort(key=lambda r: r.activation or 0, reverse=True)
-
-        return results, timings
-
-    def _convert_seeds(
-        self,
-        seeds: list[RetrievalResult] | None,
-        score_attr: str,
-    ) -> list[SeedNode]:
-        """Convert RetrievalResult seeds to SeedNode format."""
-        if not seeds:
-            return []
-
-        result = []
-        for seed in seeds:
-            score = getattr(seed, score_attr, None)
-            if score is None:
-                score = seed.activation or seed.similarity or 1.0
-            result.append(SeedNode(node_id=seed.id, score=score))
-
-        return result
-
-    async def _find_semantic_seeds(
-        self,
-        pool,
-        query_embedding_str: str,
-        bank_id: str,
-        fact_type: str,
-        limit: int = 20,
-        threshold: float = 0.3,
-        tags: list[str] | None = None,
-        tags_match: TagsMatch = "any",
-    ) -> list[SeedNode]:
-        """Fallback: find semantic seeds via embedding search."""
-        from .tags import build_tags_where_clause_simple
-
-        tags_clause = build_tags_where_clause_simple(tags, 6, match=tags_match)
-        params = [query_embedding_str, bank_id, fact_type, threshold, limit]
-        if tags:
-            params.append(tags)
-
-        async with acquire_with_retry(pool) as conn:
-            rows = await conn.fetch(
-                f"""
-                SELECT id, 1 - (embedding <=> $1::vector) AS similarity
-                FROM {fq_table("memory_units")}
-                WHERE bank_id = $2
-                  AND embedding IS NOT NULL
-                  AND fact_type = $3
-                  AND (1 - (embedding <=> $1::vector)) >= $4
-                  {tags_clause}
-                ORDER BY embedding <=> $1::vector
-                LIMIT $5
-                """,
-                *params,
-            )
-
-        return [SeedNode(node_id=str(r["id"]), score=r["similarity"]) for r in rows]

hindsight-api-slim/hindsight_api/engine/search/reranking.py

@@ -2,6 +2,7 @@
 Cross-encoder neural reranking for search results.
 """
 
+import math
 from datetime import datetime, timezone
 
 from .types import MergedCandidate, ScoredResult
@@ -13,6 +14,7 @@ UTC = timezone.utc
 # so the max combined boost is (1 + alpha/2)^2 ≈ +21% and min is (1 - alpha/2)^2 ≈ -19%.
 _RECENCY_ALPHA: float = 0.2
 _TEMPORAL_ALPHA: float = 0.2
+_PROOF_COUNT_ALPHA: float = 0.1  # Conservative: max ±5% for evidence strength
 
 
 def apply_combined_scoring(
@@ -20,32 +22,81 @@ def apply_combined_scoring(
     now: datetime,
     recency_alpha: float = _RECENCY_ALPHA,
     temporal_alpha: float = _TEMPORAL_ALPHA,
+    proof_count_alpha: float = _PROOF_COUNT_ALPHA,
+    is_passthrough_reranker: bool = False,
 ) -> None:
     """Apply combined scoring to a list of ScoredResults in-place.
 
-    Uses the cross-encoder score as the primary relevance signal, with recency
-    and temporal proximity applied as multiplicative boosts. This ensures the
-    influence of these secondary signals is always proportional to the base
-    relevance score, regardless of the cross-encoder model's score calibration.
+    Uses the cross-encoder score as the primary relevance signal, with recency,
+    temporal proximity, and proof count applied as multiplicative boosts. This
+    ensures the influence of these secondary signals is always proportional to
+    the base relevance score, regardless of the cross-encoder model's score
+    calibration.
 
     Formula::
 
-        recency_boost  = 1 + recency_alpha  * (recency  - 0.5)   # in [1-α/2, 1+α/2]
-        temporal_boost = 1 + temporal_alpha * (temporal - 0.5)   # in [1-α/2, 1+α/2]
-        combined_score = cross_encoder_score_normalized * recency_boost * temporal_boost
+        recency_boost     = 1 + recency_alpha     * (recency     - 0.5)   # in [1-α/2, 1+α/2]
+        temporal_boost    = 1 + temporal_alpha    * (temporal    - 0.5)   # in [1-α/2, 1+α/2]
+        proof_count_boost = 1 + proof_count_alpha * (proof_norm  - 0.5)   # in [1-α/2, 1+α/2]
+        combined_score    = CE_normalized * recency_boost * temporal_boost * proof_count_boost
+
+    proof_norm maps proof_count using a smooth logarithmic curve centered at 0.5,
+    clamped to [0, 1]:
+      proof_count=1 → 0.5 + 0 = 0.5 (neutral multiplier)
+      proof_count=150 → clamped to 1.0 (max +5% boost)
 
     Temporal proximity is treated as neutral (0.5) when not set by temporal retrieval,
     so temporal_boost collapses to 1.0 for non-temporal queries.
 
+    Proof count is treated as neutral (0.5) when not available (non-observation facts),
+    so proof_count_boost collapses to 1.0 for world/experience/opinion facts.
+
     Args:
         scored_results: Results from the cross-encoder reranker. Mutated in place.
         now: Current UTC datetime for recency calculation.
         recency_alpha: Max relative recency adjustment (default 0.2 → ±10%).
         temporal_alpha: Max relative temporal adjustment (default 0.2 → ±10%).
+        proof_count_alpha: Max relative proof count adjustment (default 0.1 → ±5%).
     """
     if now.tzinfo is None:
         now = now.replace(tzinfo=UTC)
 
+    # When the configured cross-encoder is a passthrough (e.g.
+    # RRFPassthroughCrossEncoder used by slim deployments), every
+    # cross_encoder_score_normalized is identical and provides no relevance
+    # signal. In that case the multiplicative recency / temporal / proof_count
+    # boosts below become the *only* ranking signal — making the final order a
+    # pure recency sort regardless of how relevant a candidate actually is.
+    #
+    # Detect that case and seed cross_encoder_score_normalized from the RRF
+    # rank instead, so the boosts modulate a meaningful base score rather than
+    # replacing it. This is a no-op for real cross-encoders, which produce
+    # diverse scores.
+    # When the reranker is a passthrough (e.g. RRFPassthroughCrossEncoder used
+    # by slim deployments), every cross_encoder_score_normalized is identical
+    # and provides no relevance signal. The multiplicative recency / temporal /
+    # proof_count boosts below would then become the *only* ranking signal,
+    # making the final order a pure recency sort regardless of how relevant a
+    # candidate actually is.
+    #
+    # Seed cross_encoder_score_normalized from the RRF rank instead, so the
+    # boosts modulate a meaningful base score. Caller passes is_passthrough
+    # explicitly because "all scores identical" is too fragile a heuristic —
+    # a real reranker can also tie scores (especially in tests with synthetic
+    # data) and we'd corrupt legitimate single-result reranks.
+    if is_passthrough_reranker and scored_results:
+        n = len(scored_results)
+        sorted_by_rrf = sorted(
+            scored_results,
+            key=lambda s: getattr(getattr(s, "candidate", None), "rrf_score", 0.0),
+            reverse=True,
+        )
+        denom = max(1, n - 1)
+        for new_rank, sr in enumerate(sorted_by_rrf):
+            # Map rank → [0.1, 1.0] so the recency boost can still nudge
+            # ordering between adjacent candidates without overpowering RRF.
+            sr.cross_encoder_score_normalized = 1.0 - (0.9 * new_rank / denom)
+
     for sr in scored_results:
         # Recency: linear decay over 365 days → [0.1, 1.0]; neutral 0.5 if no date.
         sr.recency = 0.5
@@ -59,13 +110,23 @@ def apply_combined_scoring(
         # Temporal proximity: meaningful only for temporal queries; neutral otherwise.
         sr.temporal = sr.retrieval.temporal_proximity if sr.retrieval.temporal_proximity is not None else 0.5
 
+        # Proof count: log-normalized evidence strength; neutral for non-observations.
+        proof_count = sr.retrieval.proof_count
+        if proof_count is not None and proof_count >= 1:
+            # Clamp to [0, 1] so extreme counts stay within documented ±5% range
+            proof_norm = min(1.0, max(0.0, 0.5 + (math.log(proof_count) / 10.0)))
+        else:
+            # Neutral baseline is precisely 0.5, ensuring neutral multiplier (1.0)
+            proof_norm = 0.5
+
         # RRF: kept at 0.0 for trace continuity but excluded from scoring.
         # RRF is batch-relative (min-max normalised) and redundant after reranking.
         sr.rrf_normalized = 0.0
 
         recency_boost = 1.0 + recency_alpha * (sr.recency - 0.5)
         temporal_boost = 1.0 + temporal_alpha * (sr.temporal - 0.5)
-        sr.combined_score = sr.cross_encoder_score_normalized * recency_boost * temporal_boost
+        proof_count_boost = 1.0 + proof_count_alpha * (proof_norm - 0.5)
+        sr.combined_score = sr.cross_encoder_score_normalized * recency_boost * temporal_boost * proof_count_boost
         sr.weight = sr.combined_score
 
 
@@ -153,6 +214,8 @@ class CrossEncoderReranker:
 
         # Normalize scores using sigmoid to [0, 1] range
         # Cross-encoder returns logits which can be negative
+        import math
+
         import numpy as np
 
         def sigmoid(x):
@@ -163,11 +226,20 @@ class CrossEncoderReranker:
         # Create ScoredResult objects with cross-encoder scores
         scored_results = []
         for candidate, raw_score, norm_score in zip(candidates, scores, normalized_scores):
+            # Sanitize NaN scores (cross-encoder can return NaN for certain inputs).
+            # NaN propagates through all downstream scoring and Pydantic serializes
+            # NaN as JSON null, which breaks clients expecting numeric values.
+            raw = float(raw_score)
+            norm = float(norm_score)
+            if math.isnan(raw):
+                raw = 0.0
+            if math.isnan(norm):
+                norm = 0.0
             scored_result = ScoredResult(
                 candidate=candidate,
-                cross_encoder_score=float(raw_score),
-                cross_encoder_score_normalized=float(norm_score),
-                weight=float(norm_score),  # Initial weight is just cross-encoder score
+                cross_encoder_score=raw,
+                cross_encoder_score_normalized=norm,
+                weight=norm,  # Initial weight is just cross-encoder score
             )
             scored_results.append(scored_result)
 

hindsight-api-slim/hindsight_api/engine/search/retrieval.py

@@ -10,6 +10,7 @@ Implements:
 
 import asyncio
 import logging
+import re
 from dataclasses import dataclass, field
 from datetime import UTC, datetime
 from typing import Optional
@@ -17,15 +18,23 @@ from typing import Optional
 from ...config import get_config
 from ..db_utils import acquire_with_retry
 from ..memory_engine import fq_table
-from .graph_retrieval import BFSGraphRetriever, GraphRetriever
+from .graph_retrieval import GraphRetriever
 from .link_expansion_retrieval import LinkExpansionRetriever
-from .mpfp_retrieval import MPFPGraphRetriever
-from .tags import TagsMatch, build_tags_where_clause_simple
-from .types import MPFPTimings, RetrievalResult
+from .tags import TagGroup, TagsMatch, build_tag_groups_where_clause, build_tags_where_clause_simple
+from .types import GraphRetrievalTimings, RetrievalResult
 
 logger = logging.getLogger(__name__)
 
 
+def tokenize_query(query_text: str) -> list[str]:
+    """Normalize query text and split into BM25 tokens.
+
+    Strips punctuation, lowercases, and splits on whitespace.
+    Returns an empty list when the query contains no word characters.
+    """
+    return re.sub(r"[^\w\s]", " ", query_text.lower()).split()
+
+
 @dataclass
 class ParallelRetrievalResult:
     """Result from parallel retrieval across all methods."""
@@ -36,7 +45,9 @@ class ParallelRetrievalResult:
     temporal: list[RetrievalResult] | None
     timings: dict[str, float] = field(default_factory=dict)
     temporal_constraint: tuple | None = None  # (start_date, end_date)
-    mpfp_timings: list[MPFPTimings] = field(default_factory=list)  # MPFP sub-step timings per fact type
+    graph_timings: list[GraphRetrievalTimings] = field(
+        default_factory=list
+    )  # Graph retrieval sub-step timings per fact type
     max_conn_wait: float = 0.0  # Maximum connection acquisition wait time across all methods
 
 
@@ -62,15 +73,7 @@ def get_default_graph_retriever() -> GraphRetriever:
     if _default_graph_retriever is None:
         config = get_config()
         retriever_type = config.graph_retriever.lower()
-        if retriever_type == "mpfp":
-            _default_graph_retriever = MPFPGraphRetriever()
-            logger.info(
-                f"Using MPFP graph retriever (top_k_neighbors={_default_graph_retriever.config.top_k_neighbors})"
-            )
-        elif retriever_type == "bfs":
-            _default_graph_retriever = BFSGraphRetriever()
-            logger.info("Using BFS graph retriever")
-        elif retriever_type == "link_expansion":
+        if retriever_type == "link_expansion":
             _default_graph_retriever = LinkExpansionRetriever()
             logger.info("Using LinkExpansion graph retriever")
         else:
@@ -94,6 +97,7 @@ async def retrieve_semantic_bm25_combined(
     limit: int,
     tags: list[str] | None = None,
     tags_match: TagsMatch = "any",
+    tag_groups: list[TagGroup] | None = None,
 ) -> dict[str, tuple[list[RetrievalResult], list[RetrievalResult]]]:
     """
     Combined semantic + BM25 retrieval for multiple fact types in a single query.
@@ -128,31 +132,37 @@ async def retrieve_semantic_bm25_combined(
     Returns:
         Dict mapping fact_type -> (semantic_results, bm25_results)
     """
-    import re
-
     result_dict: dict[str, tuple[list[RetrievalResult], list[RetrievalResult]]] = {ft: ([], []) for ft in fact_types}
 
-    sanitized_text = re.sub(r"[^\w\s]", " ", query_text.lower())
-    tokens = [token for token in sanitized_text.split() if token]
+    tokens = tokenize_query(query_text)
 
     # Over-fetch for HNSW approximation; semantic results trimmed to limit in Python.
     hnsw_fetch = max(limit * 5, 100)
 
     cols = (
         "id, text, context, event_date, occurred_start, occurred_end, mentioned_at, "
-        "fact_type, document_id, chunk_id, tags"
+        "fact_type, document_id, chunk_id, tags, metadata, proof_count"
     )
     table = fq_table("memory_units")
 
     # --- Parameter layout ---
     # $1 = query_emb_str  (semantic arms)
     # $2 = bank_id
-    # $3 = limit          (BM25 LIMIT; semantic uses inlined hnsw_fetch literal)
-    # $4 = bm25_text      (only when tokens present)
-    # $N = tags           (N=4 when no tokens, N=5 when tokens present)
-    tags_param_idx = 5 if tokens else 4
+    # When tokens present:
+    #   $3 = limit          (BM25 LIMIT; semantic uses inlined hnsw_fetch literal)
+    #   $4 = bm25_text
+    #   $5 = tags           (if present)
+    #   $6+ = tag_groups params (one per leaf)
+    # When no tokens ($3 is skipped — not included in params to avoid type inference gap):
+    #   $3 = tags           (if present)
+    #   $4+ = tag_groups params (one per leaf)
+    tags_param_idx = 5 if tokens else 3
     tags_clause = build_tags_where_clause_simple(tags, tags_param_idx, match=tags_match)
 
+    # tag_groups params start immediately after the tags param slot
+    tag_groups_param_start = tags_param_idx + (1 if tags else 0)
+    groups_clause, groups_params, _ = build_tag_groups_where_clause(tag_groups, tag_groups_param_start)
+
     # --- Semantic UNION ALL arms (one per fact_type) ---
     # Each arm has its own ORDER BY embedding <=> $1 LIMIT {hnsw_fetch}, which
     # lets the planner use the partial HNSW index for that fact_type.
@@ -169,6 +179,7 @@ async def retrieve_semantic_bm25_combined(
             f"   AND embedding IS NOT NULL"
             f"   AND (1 - (embedding <=> $1::vector)) >= 0.3"
             f"   {tags_clause}"
+            f"   {groups_clause}"
             f" ORDER BY embedding <=> $1::vector"
             f" LIMIT {hnsw_fetch})"
         )
@@ -208,17 +219,20 @@ async def retrieve_semantic_bm25_combined(
                 f"   AND fact_type = '{ft}'"
                 f"   {bm25_where_filter}"
                 f"   {tags_clause}"
+                f"   {groups_clause}"
                 f" ORDER BY {bm25_order_by}"
                 f" LIMIT $3)"
             )
 
     query = "\nUNION ALL\n".join(arms)
 
-    params: list = [query_emb_str, bank_id, limit]
+    params: list = [query_emb_str, bank_id]
     if tokens:
-        params.append(bm25_text_param)
+        params.append(limit)  # $3: BM25 LIMIT (only referenced when tokens are present)
+        params.append(bm25_text_param)  # $4
     if tags:
         params.append(tags)
+    params.extend(groups_params)
 
     rows = await conn.fetch(query, *params)
 
@@ -251,6 +265,7 @@ async def retrieve_temporal_combined(
     semantic_threshold: float = 0.1,
     tags: list[str] | None = None,
     tags_match: TagsMatch = "any",
+    tag_groups: list[TagGroup] | None = None,
 ) -> dict[str, list[RetrievalResult]]:
     """
     Temporal retrieval for multiple fact types in a single query.
@@ -280,10 +295,14 @@ async def retrieve_temporal_combined(
         end_date = end_date.replace(tzinfo=UTC)
 
     # Build tags clause
+    # Entry point query: fixed params are $1-$6, tags at $7
     tags_clause = build_tags_where_clause_simple(tags, 7, match=tags_match)
-    params = [query_emb_str, bank_id, fact_types, start_date, end_date, semantic_threshold]
+    tag_groups_param_start = 7 + (1 if tags else 0)
+    groups_clause, groups_params, _ = build_tag_groups_where_clause(tag_groups, tag_groups_param_start)
+    params: list = [query_emb_str, bank_id, fact_types, start_date, end_date, semantic_threshold]
     if tags:
         params.append(tags)
+    params.extend(groups_params)
 
     # Two-phase entry point query:
     # Phase 1 (date_ranked): rank by date only — no embedding computation — for all units in
@@ -314,9 +333,10 @@ async def retrieve_temporal_combined(
                   (occurred_end IS NOT NULL AND occurred_end BETWEEN $4 AND $5)
               )
               {tags_clause}
+              {groups_clause}
         ),
         sim_ranked AS (
-            SELECT mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start, mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
+            SELECT mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start, mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.proof_count, mu.document_id, mu.chunk_id, mu.tags, mu.metadata,
                    1 - (mu.embedding <=> $1::vector) AS similarity,
                    ROW_NUMBER() OVER (PARTITION BY mu.fact_type ORDER BY mu.embedding <=> $1::vector) AS sim_rn
             FROM date_ranked dr
@@ -324,7 +344,7 @@ async def retrieve_temporal_combined(
             WHERE dr.rn <= 50
               AND (1 - (mu.embedding <=> $1::vector)) >= $6
         )
-        SELECT id, text, context, event_date, occurred_start, occurred_end, mentioned_at, fact_type, document_id, chunk_id, tags, similarity
+        SELECT id, text, context, event_date, occurred_start, occurred_end, mentioned_at, fact_type, proof_count, document_id, chunk_id, tags, metadata, similarity
         FROM sim_ranked
         WHERE sim_rn <= 10
         """,
@@ -400,16 +420,21 @@ async def retrieve_temporal_combined(
 
         # Build tags clause for spreading (use param 7 since 1-6 are used)
         spreading_tags_clause = build_tags_where_clause_simple(tags, 7, table_alias="mu.", match=tags_match)
+        spreading_groups_param_start = 7 + (1 if tags else 0)
+        spreading_groups_clause, spreading_groups_params, _ = build_tag_groups_where_clause(
+            tag_groups, spreading_groups_param_start, table_alias="mu."
+        )
 
         while frontier and budget_remaining > 0 and iteration < max_iterations:
             iteration += 1
             batch_ids = frontier[:batch_size]
             frontier = frontier[batch_size:]
 
-            # $1=query_emb, $2=batch_ids, $3=fact_type, $4=threshold, $5=per_source_limit, $6=bank_id, $7=tags
+            # $1=query_emb, $2=batch_ids, $3=fact_type, $4=threshold, $5=per_source_limit, $6=bank_id, $7=tags, $M+=tag_groups
             spreading_params = [query_emb_str, batch_ids, ft, semantic_threshold, per_source_limit, bank_id]
             if tags:
                 spreading_params.append(tags)
+            spreading_params.extend(spreading_groups_params)
 
             # LATERAL join: for each source node, fetch top-K neighbors by weight using
             # the existing idx_memory_links_from_type_weight index with early-exit semantics.
@@ -417,7 +442,7 @@ async def retrieve_temporal_combined(
             # bank_id on memory_units lets the planner use idx_memory_units_bank_fact_type.
             neighbors = await conn.fetch(
                 f"""
-                SELECT src.from_unit_id, mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start, mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.document_id, mu.chunk_id, mu.tags,
+                SELECT src.from_unit_id, mu.id, mu.text, mu.context, mu.event_date, mu.occurred_start, mu.occurred_end, mu.mentioned_at, mu.fact_type, mu.document_id, mu.chunk_id, mu.tags, mu.metadata,
                        l.weight, l.link_type,
                        1 - (mu.embedding <=> $1::vector) AS similarity
                 FROM unnest($2::uuid[]) AS src(from_unit_id)
@@ -436,6 +461,7 @@ async def retrieve_temporal_combined(
                   AND mu.embedding IS NOT NULL
                   AND (1 - (mu.embedding <=> $1::vector)) >= $4
                   {spreading_tags_clause}
+                  {spreading_groups_clause}
                 """,
                 *spreading_params,
             )
@@ -509,6 +535,7 @@ async def retrieve_all_fact_types_parallel(
     graph_retriever: GraphRetriever | None = None,
     tags: list[str] | None = None,
     tags_match: TagsMatch = "any",
+    tag_groups: list[TagGroup] | None = None,
 ) -> MultiFactTypeRetrievalResult:
     """
     Optimized retrieval for multiple fact types using batched queries.
@@ -566,6 +593,7 @@ async def retrieve_all_fact_types_parallel(
             thinking_budget,
             tags=tags,
             tags_match=tags_match,
+            tag_groups=tag_groups,
         )
         semantic_bm25_time = time.time() - semantic_bm25_start
 
@@ -584,6 +612,7 @@ async def retrieve_all_fact_types_parallel(
                 semantic_threshold=0.1,
                 tags=tags,
                 tags_match=tags_match,
+                tag_groups=tag_groups,
             )
             temporal_time = time.time() - temporal_start
 
@@ -591,9 +620,11 @@ async def retrieve_all_fact_types_parallel(
     timings["temporal_combined"] = temporal_time
 
     # Step 3: Run graph retrieval for each fact type in parallel
-    async def run_graph_for_fact_type(ft: str) -> tuple[str, list[RetrievalResult], float, MPFPTimings | None]:
+    async def run_graph_for_fact_type(
+        ft: str,
+    ) -> tuple[str, list[RetrievalResult], float, GraphRetrievalTimings | None]:
         graph_start = time.time()
-        results, mpfp_timing = await retriever.retrieve(
+        results, graph_timing = await retriever.retrieve(
             pool=pool,
             query_embedding_str=query_embedding_str,
             bank_id=bank_id,
@@ -604,8 +635,9 @@ async def retrieve_all_fact_types_parallel(
             temporal_seeds=None,
             tags=tags,
             tags_match=tags_match,
+            tag_groups=tag_groups,
         )
-        return ft, results, time.time() - graph_start, mpfp_timing
+        return ft, results, time.time() - graph_start, graph_timing
 
     # Run graph for all fact types in parallel
     graph_tasks = [run_graph_for_fact_type(ft) for ft in fact_types]
@@ -614,7 +646,7 @@ async def retrieve_all_fact_types_parallel(
     # Organize results by fact type
     results_by_fact_type: dict[str, ParallelRetrievalResult] = {}
     max_conn_wait = conn_wait  # Single connection for semantic+bm25+temporal
-    all_mpfp_timings: list[MPFPTimings] = []
+    all_graph_timings: list[GraphRetrievalTimings] = []
 
     for ft in fact_types:
         # Get semantic + bm25 results for this fact type
@@ -623,14 +655,14 @@ async def retrieve_all_fact_types_parallel(
         # Find graph results for this fact type
         graph_results = []
         graph_time = 0.0
-        mpfp_timing = None
+        graph_timing = None
         for gr in graph_results_list:
             if gr[0] == ft:
                 graph_results = gr[1]
                 graph_time = gr[2]
-                mpfp_timing = gr[3]
-                if mpfp_timing:
-                    all_mpfp_timings.append(mpfp_timing)
+                graph_timing = gr[3]
+                if graph_timing:
+                    all_graph_timings.append(graph_timing)
                 break
 
         # Get temporal results for this fact type from combined result
@@ -651,7 +683,7 @@ async def retrieve_all_fact_types_parallel(
                 "temporal_extraction": temporal_extraction_time,
             },
             temporal_constraint=temporal_constraint,
-            mpfp_timings=[mpfp_timing] if mpfp_timing else [],
+            graph_timings=[graph_timing] if graph_timing else [],
             max_conn_wait=max_conn_wait,
         )
 

hindsight-api-slim/hindsight_api/engine/search/tags.py

@@ -12,7 +12,11 @@ OR matching (any/any_strict): Memory matches if ANY of its tags overlap with req
 AND matching (all/all_strict): Memory matches if ALL request tags are present in its tags
 """
 
-from typing import Literal
+from __future__ import annotations
+
+from typing import Annotated, Literal
+
+from pydantic import BaseModel, ConfigDict, Field
 
 TagsMatch = Literal["any", "all", "any_strict", "all_strict"]
 
@@ -170,3 +174,217 @@ def filter_results_by_tags(
                     filtered.append(result)
 
     return filtered
+
+
+# =============================================================================
+# Compound tag group models (recursive boolean expressions)
+# =============================================================================
+
+
+class TagGroupLeaf(BaseModel):
+    """A leaf tag filter: matches memories by tag list and match mode."""
+
+    tags: list[str]
+    match: TagsMatch = "any_strict"
+
+
+class TagGroupAnd(BaseModel):
+    """Compound AND group: all child filters must match."""
+
+    model_config = ConfigDict(populate_by_name=True)
+    filters: list[TagGroup] = Field(alias="and")
+
+
+class TagGroupOr(BaseModel):
+    """Compound OR group: at least one child filter must match."""
+
+    model_config = ConfigDict(populate_by_name=True)
+    filters: list[TagGroup] = Field(alias="or")
+
+
+class TagGroupNot(BaseModel):
+    """Compound NOT group: child filter must NOT match."""
+
+    model_config = ConfigDict(populate_by_name=True)
+    filter: TagGroup = Field(alias="not")
+
+
+# TagGroup is a discriminated union; Pydantic will try left-to-right.
+# TagGroupLeaf is identified by the presence of 'tags'.
+# TagGroupAnd / TagGroupOr / TagGroupNot are compound (no 'tags' key).
+TagGroup = Annotated[
+    TagGroupLeaf | TagGroupAnd | TagGroupOr | TagGroupNot,
+    Field(union_mode="left_to_right"),
+]
+
+# Rebuild forward-reference models so recursive TagGroup is resolved.
+TagGroupAnd.model_rebuild()
+TagGroupOr.model_rebuild()
+TagGroupNot.model_rebuild()
+
+
+# =============================================================================
+# SQL builder for compound tag groups
+# =============================================================================
+
+
+def _build_group_clause(
+    group: TagGroup,
+    param_offset: int,
+    table_alias: str,
+) -> tuple[str, list, int]:
+    """
+    Recursively build an inner SQL clause (no leading AND/OR) for a single TagGroup.
+
+    Returns:
+        (inner_clause, params, next_param_offset)
+    """
+    if isinstance(group, TagGroupLeaf):
+        column = f"{table_alias}tags" if table_alias else "tags"
+        operator, include_untagged = _parse_tags_match(group.match)
+        if include_untagged:
+            clause = f"({column} IS NULL OR {column} = '{{}}' OR {column} {operator} ${param_offset})"
+        else:
+            clause = f"({column} IS NOT NULL AND {column} != '{{}}' AND {column} {operator} ${param_offset})"
+        return clause, [group.tags], param_offset + 1
+
+    elif isinstance(group, TagGroupAnd):
+        parts = []
+        params: list = []
+        offset = param_offset
+        for child in group.filters:
+            child_clause, child_params, offset = _build_group_clause(child, offset, table_alias)
+            parts.append(child_clause)
+            params.extend(child_params)
+        inner = " AND ".join(parts)
+        return f"({inner})", params, offset
+
+    elif isinstance(group, TagGroupOr):
+        parts = []
+        params = []
+        offset = param_offset
+        for child in group.filters:
+            child_clause, child_params, offset = _build_group_clause(child, offset, table_alias)
+            parts.append(child_clause)
+            params.extend(child_params)
+        inner = " OR ".join(parts)
+        return f"({inner})", params, offset
+
+    elif isinstance(group, TagGroupNot):
+        child_clause, child_params, next_offset = _build_group_clause(group.filter, param_offset, table_alias)
+        return f"NOT {child_clause}", child_params, next_offset
+
+    else:
+        # Should never happen with proper Pydantic validation
+        return "", [], param_offset
+
+
+def build_tag_groups_where_clause(
+    tag_groups: list[TagGroup] | None,
+    param_offset: int,
+    table_alias: str = "",
+) -> tuple[str, list, int]:
+    """
+    Build a SQL WHERE clause for compound tag group filtering.
+
+    Top-level groups are AND-ed together. Each group is a recursive boolean
+    expression (leaf, and, or, not).
+
+    Args:
+        tag_groups: List of TagGroup objects. If None or empty, returns empty clause.
+        param_offset: Starting parameter number for SQL placeholders.
+        table_alias: Optional table alias prefix (e.g., "mu." for "memory_units mu").
+
+    Returns:
+        Tuple of (sql_clause, params, next_param_offset):
+        - sql_clause: SQL WHERE clause string starting with "AND" (or empty string)
+        - params: List of parameter values to bind (one per leaf node)
+        - next_param_offset: Next available parameter number
+
+    Example:
+        >>> groups = [TagGroupLeaf(tags=["user:alice"], match="all_strict")]
+        >>> clause, params, next_offset = build_tag_groups_where_clause(groups, 3)
+        >>> print(clause)  # "AND (tags IS NOT NULL AND tags != '{}' AND tags @> $3)"
+    """
+    if not tag_groups:
+        return "", [], param_offset
+
+    all_params: list = []
+    all_clauses: list[str] = []
+    offset = param_offset
+
+    for group in tag_groups:
+        inner_clause, group_params, offset = _build_group_clause(group, offset, table_alias)
+        all_clauses.append(inner_clause)
+        all_params.extend(group_params)
+
+    combined = " AND ".join(all_clauses)
+    return f"AND {combined}", all_params, offset
+
+
+# =============================================================================
+# Python-side filter for compound tag groups (post-retrieval filtering)
+# =============================================================================
+
+
+def _match_group(result: object, group: TagGroup) -> bool:
+    """
+    Recursively evaluate a TagGroup against a retrieval result.
+
+    Args:
+        result: Any object with a 'tags' attribute (list[str] or None).
+        group: The TagGroup to evaluate.
+
+    Returns:
+        True if the result matches the group, False otherwise.
+    """
+    if isinstance(group, TagGroupLeaf):
+        result_tags = getattr(result, "tags", None)
+        is_untagged = result_tags is None or len(result_tags) == 0
+        _, include_untagged = _parse_tags_match(group.match)
+        is_any_match = group.match in ("any", "any_strict")
+        tags_set = set(group.tags)
+
+        if is_untagged:
+            return include_untagged
+        else:
+            result_tags_set = set(result_tags)
+            if is_any_match:
+                return bool(result_tags_set & tags_set)
+            else:
+                return tags_set <= result_tags_set
+
+    elif isinstance(group, TagGroupAnd):
+        return all(_match_group(result, child) for child in group.filters)
+
+    elif isinstance(group, TagGroupOr):
+        return any(_match_group(result, child) for child in group.filters)
+
+    elif isinstance(group, TagGroupNot):
+        return not _match_group(result, group.filter)
+
+    else:
+        return True
+
+
+def filter_results_by_tag_groups(
+    results: list,
+    tag_groups: list[TagGroup] | None,
+) -> list:
+    """
+    Filter retrieval results by compound tag groups in Python (for post-processing).
+
+    Used when SQL filtering isn't possible (e.g., graph traversal results).
+    Top-level groups are AND-ed together.
+
+    Args:
+        results: List of RetrievalResult objects with a 'tags' attribute.
+        tag_groups: List of TagGroup objects. If None or empty, returns all results.
+
+    Returns:
+        Filtered list of results where ALL top-level groups match.
+    """
+    if not tag_groups:
+        return results
+
+    return [r for r in results if all(_match_group(r, group) for group in tag_groups)]

hindsight-api-slim/hindsight_api/engine/search/think_utils.py

@@ -62,13 +62,14 @@ def format_facts_for_prompt(facts: list[MemoryFact]) -> str:
         if fact.context:
             fact_obj["context"] = fact.context
 
-        # Add occurred_start if available (when the fact occurred)
-        if fact.occurred_start:
-            occurred_start = fact.occurred_start
-            if isinstance(occurred_start, str):
-                fact_obj["occurred_start"] = occurred_start
-            elif isinstance(occurred_start, datetime):
-                fact_obj["occurred_start"] = occurred_start.strftime("%Y-%m-%d %H:%M:%S")
+        # Add temporal fields if available
+        for field_name in ("occurred_start", "occurred_end", "mentioned_at"):
+            value = getattr(fact, field_name, None)
+            if value:
+                if isinstance(value, str):
+                    fact_obj[field_name] = value
+                elif isinstance(value, datetime):
+                    fact_obj[field_name] = value.strftime("%Y-%m-%d %H:%M:%S")
 
         formatted.append(fact_obj)
 
@@ -110,11 +111,7 @@ def build_think_prompt(
     context: str | None = None,
     entity_summaries_text: str | None = None,
 ) -> str:
-    """Build the think prompt for the LLM.
-
-    Note: opinion_facts_text parameter removed - opinions are now stored as mental models
-    and included via entity_summaries_text.
-    """
+    """Build the think prompt for the LLM."""
     disposition_desc = build_disposition_description(disposition)
 
     name_section = f"""

hindsight-api-slim/hindsight_api/engine/search/trace.py

@@ -131,7 +131,7 @@ class RetrievalResult(BaseModel):
     text: str = Field(description="Memory unit text content")
     context: str = Field(default="", description="Memory unit context")
     event_date: datetime | None = Field(default=None, description="When the memory occurred")
-    fact_type: str | None = Field(default=None, description="Fact type (world, experience, opinion)")
+    fact_type: str | None = Field(default=None, description="Fact type (world, experience)")
     score: float = Field(description="Score from this retrieval method")
     score_name: str = Field(description="Name of the score (e.g., 'similarity', 'bm25_score', 'activation')")
 
@@ -140,9 +140,7 @@ class RetrievalMethodResults(BaseModel):
     """Results from a single retrieval method."""
 
     method_name: Literal["semantic", "bm25", "graph", "temporal"] = Field(description="Name of retrieval method")
-    fact_type: str | None = Field(
-        default=None, description="Fact type this retrieval was for (world, experience, opinion)"
-    )
+    fact_type: str | None = Field(default=None, description="Fact type this retrieval was for (world, experience)")
     results: list[RetrievalResult] = Field(description="Retrieved results with ranks")
     duration_seconds: float = Field(description="Time taken for this retrieval")
     metadata: dict[str, Any] = Field(default_factory=dict, description="Method-specific metadata")

hindsight-api-slim/hindsight_api/engine/search/tracer.py

@@ -319,7 +319,7 @@ class SearchTracer:
             duration_seconds: Time taken for this retrieval
             score_field: Field name containing the score in data dict
             metadata: Optional metadata about this retrieval method
-            fact_type: Fact type this retrieval was for (world, experience, opinion)
+            fact_type: Fact type this retrieval was for (world, experience)
         """
         retrieval_results = []
         for rank, (doc_id, data) in enumerate(results, start=1):

hindsight-api-slim/hindsight_api/engine/search/types.py

@@ -11,8 +11,8 @@ from typing import Any
 
 
 @dataclass
-class MPFPTimings:
-    """Timing breakdown for a single MPFP retrieval call."""
+class GraphRetrievalTimings:
+    """Timing breakdown for a single graph retrieval call."""
 
     fact_type: str
     edge_count: int = 0  # Total edges loaded
@@ -47,6 +47,8 @@ class RetrievalResult:
     document_id: str | None = None
     chunk_id: str | None = None
     tags: list[str] | None = None  # Visibility scope tags
+    metadata: dict[str, str] | None = None  # User-provided metadata
+    proof_count: int | None = None  # Number of supporting memories (observations only)
 
     # Retrieval-specific scores (only one will be set depending on retrieval method)
     similarity: float | None = None  # Semantic retrieval
@@ -70,6 +72,8 @@ class RetrievalResult:
             document_id=row.get("document_id"),
             chunk_id=row.get("chunk_id"),
             tags=row.get("tags"),
+            metadata=row.get("metadata"),
+            proof_count=row.get("proof_count"),
             similarity=row.get("similarity"),
             bm25_score=row.get("bm25_score"),
             activation=row.get("activation"),
@@ -153,6 +157,7 @@ class ScoredResult:
             "document_id": self.retrieval.document_id,
             "chunk_id": self.retrieval.chunk_id,
             "tags": self.retrieval.tags,
+            "metadata": self.retrieval.metadata,
             "semantic_similarity": self.retrieval.similarity,
             "bm25_score": self.retrieval.bm25_score,
         }

hindsight-api-slim/hindsight_api/engine/task_backend.py

@@ -82,20 +82,16 @@ class TaskBackend(ABC):
 
         Args:
             task_dict: Task dictionary to execute
+
+        Raises:
+            Exception: Re-raised from executor on failure.
         """
         if self._executor is None:
             task_type = task_dict.get("type", "unknown")
             logger.warning(f"No executor registered, skipping task {task_type}")
             return
 
-        try:
-            await self._executor(task_dict)
-        except Exception as e:
-            task_type = task_dict.get("type", "unknown")
-            logger.error(f"Error executing task {task_type}: {e}")
-            import traceback
-
-            traceback.print_exc()
+        await self._executor(task_dict)
 
 
 class SyncTaskBackend(TaskBackend):