fixes

* chore: remove dead code

* chore: remove extract_opinions from test and regenerate openapi

- Remove extract_opinions parameter from test_fact_extraction_analysis
- Regenerate OpenAPI spec after removing entity observations code

* chore: update generated files and apply formatting

- Regenerate Python and TypeScript client SDKs after main merge
- Apply ruff formatting to llm_wrapper.py

* fix: accept and filter deprecated 'opinion' fact type in recall

The dead code removal eliminated support for the 'opinion' fact type,
but existing clients may still pass it. Instead of rejecting it with
a ValueError, silently filter it out before validation to maintain
backward compatibility.

.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
+# Supported providers: openai, groq, ollama, gemini, anthropic, lmstudio, vertexai
 HINDSIGHT_API_LLM_PROVIDER=openai
 HINDSIGHT_API_LLM_API_KEY=your-api-key-here
 HINDSIGHT_API_LLM_MODEL=o3-mini
@@ -13,6 +13,13 @@ HINDSIGHT_API_LLM_BASE_URL=https://api.openai.com/v1
 # HINDSIGHT_API_LLM_API_KEY=your-anthropic-api-key
 # HINDSIGHT_API_LLM_MODEL=claude-sonnet-4-20250514
 
+# Example: Google Vertex AI configuration
+# HINDSIGHT_API_LLM_PROVIDER=vertexai
+# HINDSIGHT_API_LLM_MODEL=google/gemini-2.0-flash-001
+# HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID=your-gcp-project-id
+# 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: LM Studio local configuration (Qwen 2.5 32B recommended)
 # HINDSIGHT_API_LLM_PROVIDER=lmstudio
 # HINDSIGHT_API_LLM_API_KEY=lmstudio
@@ -26,6 +33,7 @@ HINDSIGHT_API_LOG_LEVEL=info
 
 # Database (Optional - uses embedded pg0 by default)
 # HINDSIGHT_API_DATABASE_URL=postgresql://user:pass@host:5432/db
+# HINDSIGHT_API_DATABASE_SCHEMA=public  # PostgreSQL schema name (default: public)
 
 # Embeddings Configuration (Optional - uses local by default)
 # Provider: "local" (default) or "tei" (HuggingFace Text Embeddings Inference)

.github/workflows/release.yml

@@ -139,6 +139,55 @@ jobs:
         path: hindsight-clients/typescript/*.tgz
         retention-days: 1
 
+  release-moltbot-integration:
+    runs-on: ubuntu-latest
+    environment: npm
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Node.js
+      uses: actions/setup-node@v4
+      with:
+        node-version: '22'
+        registry-url: 'https://registry.npmjs.org'
+
+    - name: Install dependencies
+      working-directory: ./hindsight-integrations/moltbot
+      run: npm ci
+
+    - name: Build
+      working-directory: ./hindsight-integrations/moltbot
+      run: npm run build
+
+    - name: Publish to npm
+      working-directory: ./hindsight-integrations/moltbot
+      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/moltbot
+      run: npm pack
+
+    - name: Upload artifacts
+      uses: actions/upload-artifact@v4
+      with:
+        name: moltbot-integration
+        path: hindsight-integrations/moltbot/*.tgz
+        retention-days: 1
+
   release-control-plane:
     runs-on: ubuntu-latest
     environment: npm
@@ -366,7 +415,7 @@ jobs:
 
   create-github-release:
     runs-on: ubuntu-latest
-    needs: [release-python-packages, release-typescript-client, release-control-plane, release-rust-cli, release-docker-images, release-helm-chart]
+    needs: [release-python-packages, release-typescript-client, release-moltbot-integration, release-control-plane, release-rust-cli, release-docker-images, release-helm-chart]
     permissions:
       contents: write
 
@@ -389,6 +438,12 @@ jobs:
         name: typescript-client
         path: ./artifacts/typescript-client
 
+    - name: Download Moltbot Integration
+      uses: actions/download-artifact@v4
+      with:
+        name: moltbot-integration
+        path: ./artifacts/moltbot-integration
+
     - name: Download Control Plane
       uses: actions/download-artifact@v4
       with:
@@ -430,6 +485,8 @@ jobs:
         cp artifacts/python-packages/hindsight-embed/dist/* release-assets/ || true
         # TypeScript client
         cp artifacts/typescript-client/*.tgz release-assets/ || true
+        # Moltbot Integration
+        cp artifacts/moltbot-integration/*.tgz release-assets/ || true
         # Control Plane
         cp artifacts/control-plane/*.tgz release-assets/ || true
         # Rust CLI binaries

.github/workflows/test.yml

@@ -82,6 +82,29 @@ jobs:
     - name: Build TypeScript client
       run: npm run build --workspace=hindsight-clients/typescript
 
+  build-moltbot-integration:
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Node.js
+      uses: actions/setup-node@v4
+      with:
+        node-version: '22'
+
+    - name: Install dependencies
+      working-directory: ./hindsight-integrations/moltbot
+      run: npm ci
+
+    - name: Run tests
+      working-directory: ./hindsight-integrations/moltbot
+      run: npm test
+
+    - name: Build
+      working-directory: ./hindsight-integrations/moltbot
+      run: npm run build
+
   build-control-plane:
     runs-on: ubuntu-latest
 

.gitignore

@@ -45,9 +45,12 @@ hindsight-docs/static/llms-full.txt
 
 hindsight-dev/benchmarks/locomo/results/
 hindsight-dev/benchmarks/longmemeval/results/
+hindsight-dev/benchmarks/consolidation/results/
+benchmarks/results/
 hindsight-cli/target
 hindsight-clients/rust/target
 .claude
 whats-next.md
 TASK.md
-CHANGELOG.md
+# Changelog is now tracked in hindsight-docs/src/pages/changelog.md
+# CHANGELOG.md

README.md

@@ -1,6 +1,6 @@
 <div align="center">
 
-![Hindsight Banner](./hindsight-docs/static/img/banner.svg)
+![Hindsight Banner](./hindsight-docs/static/img/hindsight-github-banner.png)
 
 [Documentation](https://hindsight.vectorize.io) • [Paper](https://arxiv.org/abs/2512.12818) • [Cookbook](https://hindsight.vectorize.io/cookbook) • [Hindsight Cloud](https://vectorize.io/hindsight/cloud)
 
@@ -17,55 +17,31 @@
 
 ## What is Hindsight?
 
-Hindsight™ is an agent memory system built to create smarter agents that learn over time. It eliminates the shortcomings of alternative techniques such as RAG and knowledge graph and delivers state-of-the-art performance on long term memory tasks.
+Hindsight™ is an agent memory system built to create smarter agents that learn over time. Most agent memory systems focus on recalling conversation history. Hindsight is focused on making agents that learn, not just remember.
 
-Hindsight addresses common challenges that have frustrated AI engineers building agents to automate tasks and assist users with conversational interfaces. Many of these challenges stem directly from a lack of memory.
 
-- **Inconsistency:** Agents complete tasks successfully one time, then fail when asked to complete the same task again. Memory gives the agent a mechanism to remember what worked and what didn't and to use that information to reduce errors and improve consistency.
-- **Hallucinations:** Long term memory can be seeded with external knowledge to ground agent behavior in reliable sources to augment training data.
-- **Cognitive Overload:** As workflows get complex, retrievals, tool calls, user messages and agent responses can grow to fill the context window leading to context rot. Short term memory optimization allows agents to reduce tokens and focus context by removing irrelevant details.
+<video src="https://github.com/user-attachments/assets/923b798d-3581-4897-bb62-9cfa5a931682" controls></video>
 
-## How is Hindsight Different From Other Memory Systems?
-
-![Overview](./hindsight-docs/static/img/hindsight-overview.webp)
-
-Most agent memory implementation rely on basic vector search or sometimes use a knowledge graph. Hindsight uses biomimetic data structures to organize agent memories in a way that is more like how human memory works:
-
-- **World:** Facts about the world ("The stove gets hot")
-- **Experiences:** Agent's own experiences ("I touched the stove and it really hurt")
-- **Opinion:** Beliefs with confidence scores ("I shouldn't touch the stove again" - .99 confidence)
-- **Observation:** Complex mental models derived by reflecting on facts and experiences ("Curling irons, ovens, and fire are also hot. I shouldn't touch those either.")
-
-Memories in Hindsight are stored in banks (i.e. memory banks). When memories are added to Hindsight, they are pushed into either the world facts or experiences memory pathway. They are then represented as a combination of entities, relationships, and time series with sparse/dense vector representations to aid in later recall.
-
-Hindsight provides three simple methods to interact with the system:
-
-- **Retain:** Provide information to Hindsight that you want it to remember
-- **Recall:** Retrieve memories from Hindsight
-- **Reflect:** Reflect on memories and experiences to generate new observations and insights from existing memories.
-
-### Agent Memory That Learns
-
-A key goal of Hindsight is to build agent memory that enables agents to learn and improve over time. This is the role of the `reflect` operation which provides the agent to form broader opinions and observations over time.
-
-For example, imagine a product support agent that is helping a user troubleshoot a problem. It uses a `search-documentation` tool it found on an MCP server. Later in the conversation, the agent discovers that the documentation returned from the tool wasn't for the product the user was asking about. The agent now has an experience in its memory bank. And just like humans, we want that agent to learn from its experience.
-
-As the agent gains more experiences, `reflect` allows the agent to form observations about what worked, what didn't, and what to do differently the next time it encounters a similar task.
-
----
+It eliminates the shortcomings of alternative techniques such as RAG and knowledge graph and delivers state-of-the-art performance on long term memory tasks.
 
 ## Memory Performance & Accuracy
 
-Hindsight has achieved state-of-the-art performance on the LongMemEval benchmark, widely used to assess memory system performance across a variety of conversational
-AI scenarios. The current reported performance of Hindsight and other agent memory solutions as of December 2025 is shown here:
+Hindsight is the most accurate agent memory system ever tested according to benchmark performance. It has achieved state-of-the-art performance on the LongMemEval benchmark, widely used to assess memory system performance across a variety of conversational AI scenarios. The current reported performance of Hindsight and other agent memory solutions as of January 2026 is shown here:
 
 ![Overview](./hindsight-docs/static/img/hindsight-bench.jpg)
 
-The benchmark performance data for Hindsight and GPT-4o (full context) have been reproduced by research collaborators at the Virginia Tech [Sanghani Center for Artificial Intelligence and Data Analytics](https://sanghani.cs.vt.edu/) and The Washington Post. Other scores are self-reported by software vendors.
+The benchmark performance data for Hindsight has been independently reproduced by research collaborators at the Virginia Tech [Sanghani Center for Artificial Intelligence and Data Analytics](https://sanghani.cs.vt.edu/) and The Washington Post. Other scores are self-reported by software vendors.
 
-A thorough examination of the techniques implemented in Hindsight and detailed breakdowns of benchmark performance are [available on arXiv](https://arxiv.org/abs/2512.12818). This research is currently being prepared for conference submission and the wider peer review process.
+Hindsight is being used in production at Fortune 500 enterprises and by a growing number of AI startups. 
+
+## Adding Hindsight to Your AI Agents
+
+The easiest way use Hindsight with an existing agent is with the LLM Wrapper. You can add memory to your agent with 2 lines of code. That will swap your current LLM client out with the Hindsight wrapper. After that, memories will be stored and retrieved automatically as you make LLM calls.
+
+If you need more control over how and when your agent stores and recalls memories, there's also a simple API you can integrate with using the SDKs or directly via HTTP.
+
+![Hindsight Banner](./hindsight-docs/static/img/migration-code.png)
 
-The benchmark results from this research can be inspected in our [visual benchmark explorer](https://hindsight-benchmarks.vercel.app). As additional improvements are made to Hindsight, new benchmark data will be available for review using this same tool.
 
 ## Quick Start
 
@@ -148,8 +124,45 @@ await client.recall('my-bank', 'What does Alice like?');
 
 ---
 
+## Use Cases
+
+
+Hindsight is built to support conversational AI agents as well as agents that are intended to perform tasks autonomously. The ideal use case for Hindsight are agents that require a blend of these features such as AI employees that need to handle open-ended tasks, change behavior based on user feedback, and learn to perform complex tasks to automate work at a level that approximates a human work. Hindsight can be used with simple AI workflows like those built with n8n and other similar tools, but may be overkill for such applications.
+
+### Per-User Memories and Chat History
+
+One of the simpler use cases you can use Hindsight for is to personalize AI chatbots and other conversational agents by storing and recalling memories associated with individual users.
+
+The requirements for this use case usually look something like this:
+
+![Per-User Memories](./hindsight-docs/static/img/per-user-memory-requirements.png)
+
+<video src="https://github.com/user-attachments/assets/4805e8e1-e7d1-47c6-a4f8-2344a5ec8906" controls></video>
+
+Satisfying these requirements in Hindsight is straightforward. When new user inputs and tool calls are ingested into Hindsight using the retain operation, custom metadata can be used to enrich the new memories. Metadata provides a convenient way to isolate memories that need to be restricted to a given user. Once these are fed into the retain operation, any raw memories and mental models that get created can be filtered when retrieving relevant memories. 
+
+![Per-User Memories](./hindsight-docs/static/img/per-user-memory-howto.png)
+
+---
+
 ## Architecture & Operations
 
+![Overview](./hindsight-docs/static/img/hindsight-overview.webp)
+
+Most agent memory implementation rely on basic vector search or sometimes use a knowledge graph. Hindsight uses biomimetic data structures to organize agent memories in a way that is more like how human memory works:
+
+- **World:** Facts about the world ("The stove gets hot")
+- **Experiences:** Agent's own experiences ("I touched the stove and it really hurt")
+- **Mental Models:** Learned understanding of the agent's world formed by reflecting on raw memories and experiences.
+
+Memories in Hindsight are stored in banks (i.e. memory banks). When memories are added to Hindsight, they are pushed into either the world facts or experiences memory pathway. They are then represented as a combination of entities, relationships, and time series with sparse/dense vector representations to aid in later recall.
+
+Hindsight provides three simple methods to interact with the system:
+
+- **Retain:** Provide information to Hindsight that you want it to remember
+- **Recall:** Retrieve memories from Hindsight
+- **Reflect:** Reflect on memories and experiences to generate new observations and insights from existing memories.
+
 ### Retain
 
 The `retain` operation is used to push new memories into Hindsight. It tells Hindsight to _retain_ the information you pass in as an input.
@@ -208,7 +221,7 @@ The final output is trimmed as needed to fit within the token limit.
 
 ### Reflect
 
-The reflect operation is used to perform a more thorough analysis of existing memories. This allows the agent to form new connections between memories which are then persisted as opinions and/or observations. When building agents, the reflect operation is a key capability to enable the agent to learn from its experiences. 
+The reflect operation is used to perform a more thorough analysis of existing memories. This allows the agent to form new connections between memories and build a more thorough understanding of its world.
 
 For example, the `reflect` operation can be used to support use cases such as:
 

helm/hindsight/Chart.yaml

@@ -2,8 +2,8 @@ apiVersion: v2
 name: hindsight
 description: Hindsight helm chart
 type: application
-version: 0.3.0
-appVersion: "0.3.0"
+version: 0.4.2
+appVersion: "0.4.2"
 keywords:
   - ai
   - memory

hindsight-api/hindsight_api/__init__.py

@@ -46,4 +46,4 @@ __all__ = [
     "RemoteTEICrossEncoder",
     "LLMConfig",
 ]
-__version__ = "0.1.0"
+__version__ = "0.4.2"

hindsight-api/hindsight_api/api/http.py

@@ -92,8 +92,7 @@ class RecallRequest(BaseModel):
     query: str
     types: list[str] | None = Field(
         default=None,
-        description="List of fact types to recall: 'world', 'experience', 'observation'. Defaults to world and experience if not specified. "
-        "Note: 'opinion' is accepted but ignored (opinions are excluded from recall).",
+        description="List of fact types to recall: 'world', 'experience', 'observation'. Defaults to world and experience if not specified.",
     )
     budget: Budget = Budget.MID
     max_tokens: int = 4096
@@ -504,13 +503,6 @@ class ReflectRequest(BaseModel):
     )
 
 
-class OpinionItem(BaseModel):
-    """Model for an opinion with confidence score."""
-
-    text: str
-    confidence: float
-
-
 class ReflectFact(BaseModel):
     """A fact used in think response."""
 
@@ -529,7 +521,7 @@ class ReflectFact(BaseModel):
 
     id: str | None = None
     text: str
-    type: str | None = None  # fact type: world, experience, opinion
+    type: str | None = None  # fact type: world, experience, observation
     context: str | None = None
     occurred_start: str | None = None
     occurred_end: str | None = None
@@ -1323,7 +1315,7 @@ class VersionResponse(BaseModel):
     model_config = ConfigDict(
         json_schema_extra={
             "example": {
-                "api_version": "1.0.0",
+                "api_version": "0.4.0",
                 "features": {
                     "observations": False,
                     "mcp": True,
@@ -1412,9 +1404,10 @@ def create_app(
                 worker_id=worker_id,
                 executor=memory.execute_task,
                 poll_interval_ms=config.worker_poll_interval_ms,
-                batch_size=config.worker_batch_size,
                 max_retries=config.worker_max_retries,
                 tenant_extension=getattr(memory, "_tenant_extension", None),
+                max_slots=config.worker_max_slots,
+                consolidation_max_slots=config.worker_consolidation_max_slots,
             )
             poller_task = asyncio.create_task(poller.run())
             logging.info(f"Worker poller started (worker_id={worker_id})")
@@ -1567,11 +1560,12 @@ def _register_routes(app: FastAPI):
         Returns version info and feature flags that can be used by clients
         to determine which capabilities are available.
         """
+        from hindsight_api import __version__
         from hindsight_api.config import get_config
 
         config = get_config()
         return VersionResponse(
-            api_version="1.0.0",
+            api_version=__version__,
             features=FeaturesInfo(
                 observations=config.enable_observations,
                 mcp=config.mcp_enabled,
@@ -1706,9 +1700,7 @@ def _register_routes(app: FastAPI):
         description="Recall memory using semantic similarity and spreading activation.\n\n"
         "The type parameter is optional and must be one of:\n"
         "- `world`: General knowledge about people, places, events, and things that happen\n"
-        "- `experience`: Memories about experience, conversations, actions taken, and tasks performed\n"
-        "- `opinion`: The bank's formed beliefs, perspectives, and viewpoints\n\n"
-        "Set `include_entities=true` to get entity observations alongside recall results.",
+        "- `experience`: Memories about experience, conversations, actions taken, and tasks performed",
         operation_id="recall_memories",
         tags=["Memory"],
     )
@@ -1722,10 +1714,8 @@ def _register_routes(app: FastAPI):
         metrics = get_metrics_collector()
 
         try:
-            # Default to world and experience if not specified (exclude observation and opinion)
-            # Filter out 'opinion' even if requested - opinions are excluded from recall
+            # Default to world and experience if not specified (exclude observation)
             fact_types = request.types if request.types else list(VALID_RECALL_FACT_TYPES)
-            fact_types = [ft for ft in fact_types if ft != "opinion"]
 
             # Parse query_timestamp if provided
             question_date = None
@@ -1857,8 +1847,7 @@ def _register_routes(app: FastAPI):
         "2. Retrieves world facts relevant to the query\n"
         "3. Retrieves existing opinions (bank's perspectives)\n"
         "4. Uses LLM to formulate a contextual answer\n"
-        "5. Extracts and stores any new opinions formed\n"
-        "6. Returns plain text answer, the facts used, and new opinions",
+        "5. Returns plain text answer and the facts used",
         operation_id="reflect",
         tags=["Memory"],
     )

hindsight-api/hindsight_api/api/mcp.py

@@ -29,15 +29,26 @@ logger = logging.getLogger(__name__)
 # Default bank_id from environment variable
 DEFAULT_BANK_ID = os.environ.get("HINDSIGHT_MCP_BANK_ID", "default")
 
+# MCP authentication token (optional - if set, Bearer token auth is required)
+MCP_AUTH_TOKEN = os.environ.get("HINDSIGHT_API_MCP_AUTH_TOKEN")
+
 # Context variable to hold the current bank_id
 _current_bank_id: ContextVar[str | None] = ContextVar("current_bank_id", default=None)
 
+# Context variable to hold the current API key (for tenant auth propagation)
+_current_api_key: ContextVar[str | None] = ContextVar("current_api_key", default=None)
+
 
 def get_current_bank_id() -> str | None:
     """Get the current bank_id from context."""
     return _current_bank_id.get()
 
 
+def get_current_api_key() -> str | None:
+    """Get the current API key from context."""
+    return _current_api_key.get()
+
+
 def create_mcp_server(memory: MemoryEngine) -> FastMCP:
     """
     Create and configure the Hindsight MCP server.
@@ -54,6 +65,7 @@ def create_mcp_server(memory: MemoryEngine) -> FastMCP:
     # Configure and register tools using shared module
     config = MCPToolsConfig(
         bank_id_resolver=get_current_bank_id,
+        api_key_resolver=get_current_api_key,  # Propagate API key for tenant auth
         include_bank_id_param=True,  # HTTP MCP supports multi-bank via parameter
         tools=None,  # All tools
         retain_fire_and_forget=False,  # HTTP MCP supports sync/async modes
@@ -65,7 +77,11 @@ def create_mcp_server(memory: MemoryEngine) -> FastMCP:
 
 
 class MCPMiddleware:
-    """ASGI middleware that extracts bank_id from header or path and sets context.
+    """ASGI middleware that handles authentication and extracts bank_id from header or path.
+
+    Authentication:
+        If HINDSIGHT_API_MCP_AUTH_TOKEN is set, all requests must include a valid
+        Authorization header with Bearer token or direct token matching the configured value.
 
     Bank ID can be provided via:
     1. X-Bank-Id header (recommended for Claude Code)
@@ -74,7 +90,7 @@ class MCPMiddleware:
 
     For Claude Code, configure with:
         claude mcp add --transport http hindsight http://localhost:8888/mcp \\
-            --header "X-Bank-Id: my-bank"
+            --header "X-Bank-Id: my-bank" --header "Authorization: Bearer <token>"
     """
 
     def __init__(self, app, memory: MemoryEngine):
@@ -98,6 +114,22 @@ class MCPMiddleware:
             await self.mcp_app(scope, receive, send)
             return
 
+        # Extract auth token from header (for tenant auth propagation)
+        auth_header = self._get_header(scope, "Authorization")
+        auth_token: str | None = None
+        if auth_header:
+            # Support both "Bearer <token>" and direct token
+            auth_token = auth_header[7:].strip() if auth_header.startswith("Bearer ") else auth_header.strip()
+
+        # Authenticate if MCP_AUTH_TOKEN is configured
+        if MCP_AUTH_TOKEN:
+            if not auth_token:
+                await self._send_error(send, 401, "Authorization header required")
+                return
+            if auth_token != MCP_AUTH_TOKEN:
+                await self._send_error(send, 401, "Invalid authentication token")
+                return
+
         path = scope.get("path", "")
 
         # Strip any mount prefix (e.g., /mcp) that FastAPI might not have stripped
@@ -132,8 +164,10 @@ class MCPMiddleware:
             bank_id = DEFAULT_BANK_ID
             logger.debug(f"Using default bank_id: {bank_id}")
 
-        # Set bank_id context
-        token = _current_bank_id.set(bank_id)
+        # Set bank_id and api_key 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
         try:
             new_scope = scope.copy()
             new_scope["path"] = new_path
@@ -152,7 +186,9 @@ class MCPMiddleware:
 
             await self.mcp_app(new_scope, receive, send_wrapper)
         finally:
-            _current_bank_id.reset(token)
+            _current_bank_id.reset(bank_id_token)
+            if api_key_token is not None:
+                _current_api_key.reset(api_key_token)
 
     async def _send_error(self, send, status: int, message: str):
         """Send an error response."""
@@ -176,6 +212,10 @@ def create_mcp_app(memory: MemoryEngine):
     """
     Create an ASGI app that handles MCP requests.
 
+    Authentication:
+        Set HINDSIGHT_API_MCP_AUTH_TOKEN to require Bearer token authentication.
+        If not set, MCP endpoint is open (for local development).
+
     Bank ID can be provided via:
     1. X-Bank-Id header: claude mcp add --transport http hindsight http://localhost:8888/mcp --header "X-Bank-Id: my-bank"
     2. URL path: /mcp/{bank_id}/

hindsight-api/hindsight_api/config.py

@@ -20,11 +20,15 @@ logger = logging.getLogger(__name__)
 
 # Environment variable names
 ENV_DATABASE_URL = "HINDSIGHT_API_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"
 ENV_LLM_MODEL = "HINDSIGHT_API_LLM_MODEL"
 ENV_LLM_BASE_URL = "HINDSIGHT_API_LLM_BASE_URL"
 ENV_LLM_MAX_CONCURRENT = "HINDSIGHT_API_LLM_MAX_CONCURRENT"
+ENV_LLM_MAX_RETRIES = "HINDSIGHT_API_LLM_MAX_RETRIES"
+ENV_LLM_INITIAL_BACKOFF = "HINDSIGHT_API_LLM_INITIAL_BACKOFF"
+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"
 
@@ -33,19 +37,35 @@ ENV_RETAIN_LLM_PROVIDER = "HINDSIGHT_API_RETAIN_LLM_PROVIDER"
 ENV_RETAIN_LLM_API_KEY = "HINDSIGHT_API_RETAIN_LLM_API_KEY"
 ENV_RETAIN_LLM_MODEL = "HINDSIGHT_API_RETAIN_LLM_MODEL"
 ENV_RETAIN_LLM_BASE_URL = "HINDSIGHT_API_RETAIN_LLM_BASE_URL"
+ENV_RETAIN_LLM_MAX_CONCURRENT = "HINDSIGHT_API_RETAIN_LLM_MAX_CONCURRENT"
+ENV_RETAIN_LLM_MAX_RETRIES = "HINDSIGHT_API_RETAIN_LLM_MAX_RETRIES"
+ENV_RETAIN_LLM_INITIAL_BACKOFF = "HINDSIGHT_API_RETAIN_LLM_INITIAL_BACKOFF"
+ENV_RETAIN_LLM_MAX_BACKOFF = "HINDSIGHT_API_RETAIN_LLM_MAX_BACKOFF"
+ENV_RETAIN_LLM_TIMEOUT = "HINDSIGHT_API_RETAIN_LLM_TIMEOUT"
 
 ENV_REFLECT_LLM_PROVIDER = "HINDSIGHT_API_REFLECT_LLM_PROVIDER"
 ENV_REFLECT_LLM_API_KEY = "HINDSIGHT_API_REFLECT_LLM_API_KEY"
 ENV_REFLECT_LLM_MODEL = "HINDSIGHT_API_REFLECT_LLM_MODEL"
 ENV_REFLECT_LLM_BASE_URL = "HINDSIGHT_API_REFLECT_LLM_BASE_URL"
+ENV_REFLECT_LLM_MAX_CONCURRENT = "HINDSIGHT_API_REFLECT_LLM_MAX_CONCURRENT"
+ENV_REFLECT_LLM_MAX_RETRIES = "HINDSIGHT_API_REFLECT_LLM_MAX_RETRIES"
+ENV_REFLECT_LLM_INITIAL_BACKOFF = "HINDSIGHT_API_REFLECT_LLM_INITIAL_BACKOFF"
+ENV_REFLECT_LLM_MAX_BACKOFF = "HINDSIGHT_API_REFLECT_LLM_MAX_BACKOFF"
+ENV_REFLECT_LLM_TIMEOUT = "HINDSIGHT_API_REFLECT_LLM_TIMEOUT"
 
 ENV_CONSOLIDATION_LLM_PROVIDER = "HINDSIGHT_API_CONSOLIDATION_LLM_PROVIDER"
 ENV_CONSOLIDATION_LLM_API_KEY = "HINDSIGHT_API_CONSOLIDATION_LLM_API_KEY"
 ENV_CONSOLIDATION_LLM_MODEL = "HINDSIGHT_API_CONSOLIDATION_LLM_MODEL"
 ENV_CONSOLIDATION_LLM_BASE_URL = "HINDSIGHT_API_CONSOLIDATION_LLM_BASE_URL"
+ENV_CONSOLIDATION_LLM_MAX_CONCURRENT = "HINDSIGHT_API_CONSOLIDATION_LLM_MAX_CONCURRENT"
+ENV_CONSOLIDATION_LLM_MAX_RETRIES = "HINDSIGHT_API_CONSOLIDATION_LLM_MAX_RETRIES"
+ENV_CONSOLIDATION_LLM_INITIAL_BACKOFF = "HINDSIGHT_API_CONSOLIDATION_LLM_INITIAL_BACKOFF"
+ENV_CONSOLIDATION_LLM_MAX_BACKOFF = "HINDSIGHT_API_CONSOLIDATION_LLM_MAX_BACKOFF"
+ENV_CONSOLIDATION_LLM_TIMEOUT = "HINDSIGHT_API_CONSOLIDATION_LLM_TIMEOUT"
 
 ENV_EMBEDDINGS_PROVIDER = "HINDSIGHT_API_EMBEDDINGS_PROVIDER"
 ENV_EMBEDDINGS_LOCAL_MODEL = "HINDSIGHT_API_EMBEDDINGS_LOCAL_MODEL"
+ENV_EMBEDDINGS_LOCAL_FORCE_CPU = "HINDSIGHT_API_EMBEDDINGS_LOCAL_FORCE_CPU"
 ENV_EMBEDDINGS_TEI_URL = "HINDSIGHT_API_EMBEDDINGS_TEI_URL"
 ENV_EMBEDDINGS_OPENAI_API_KEY = "HINDSIGHT_API_EMBEDDINGS_OPENAI_API_KEY"
 ENV_EMBEDDINGS_OPENAI_MODEL = "HINDSIGHT_API_EMBEDDINGS_OPENAI_MODEL"
@@ -65,6 +85,7 @@ ENV_RERANKER_LITELLM_MODEL = "HINDSIGHT_API_RERANKER_LITELLM_MODEL"
 
 ENV_RERANKER_PROVIDER = "HINDSIGHT_API_RERANKER_PROVIDER"
 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_TEI_URL = "HINDSIGHT_API_RERANKER_TEI_URL"
 ENV_RERANKER_TEI_BATCH_SIZE = "HINDSIGHT_API_RERANKER_TEI_BATCH_SIZE"
@@ -87,17 +108,22 @@ ENV_MCP_LOCAL_BANK_ID = "HINDSIGHT_API_MCP_LOCAL_BANK_ID"
 ENV_MCP_INSTRUCTIONS = "HINDSIGHT_API_MCP_INSTRUCTIONS"
 ENV_MENTAL_MODEL_REFRESH_CONCURRENCY = "HINDSIGHT_API_MENTAL_MODEL_REFRESH_CONCURRENCY"
 
+# Vertex AI configuration
+ENV_LLM_VERTEXAI_PROJECT_ID = "HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID"
+ENV_LLM_VERTEXAI_REGION = "HINDSIGHT_API_LLM_VERTEXAI_REGION"
+ENV_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY = "HINDSIGHT_API_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY"
+
 # Retain settings
 ENV_RETAIN_MAX_COMPLETION_TOKENS = "HINDSIGHT_API_RETAIN_MAX_COMPLETION_TOKENS"
 ENV_RETAIN_CHUNK_SIZE = "HINDSIGHT_API_RETAIN_CHUNK_SIZE"
 ENV_RETAIN_EXTRACT_CAUSAL_LINKS = "HINDSIGHT_API_RETAIN_EXTRACT_CAUSAL_LINKS"
 ENV_RETAIN_EXTRACTION_MODE = "HINDSIGHT_API_RETAIN_EXTRACTION_MODE"
 ENV_RETAIN_CUSTOM_INSTRUCTIONS = "HINDSIGHT_API_RETAIN_CUSTOM_INSTRUCTIONS"
-ENV_RETAIN_OBSERVATIONS_ASYNC = "HINDSIGHT_API_RETAIN_OBSERVATIONS_ASYNC"
 
 # Observations settings (consolidated knowledge from facts)
 ENV_ENABLE_OBSERVATIONS = "HINDSIGHT_API_ENABLE_OBSERVATIONS"
 ENV_CONSOLIDATION_BATCH_SIZE = "HINDSIGHT_API_CONSOLIDATION_BATCH_SIZE"
+ENV_CONSOLIDATION_MAX_TOKENS = "HINDSIGHT_API_CONSOLIDATION_MAX_TOKENS"
 
 # Optimization flags
 ENV_SKIP_LLM_VERIFICATION = "HINDSIGHT_API_SKIP_LLM_VERIFICATION"
@@ -117,26 +143,38 @@ ENV_WORKER_ENABLED = "HINDSIGHT_API_WORKER_ENABLED"
 ENV_WORKER_ID = "HINDSIGHT_API_WORKER_ID"
 ENV_WORKER_POLL_INTERVAL_MS = "HINDSIGHT_API_WORKER_POLL_INTERVAL_MS"
 ENV_WORKER_MAX_RETRIES = "HINDSIGHT_API_WORKER_MAX_RETRIES"
-ENV_WORKER_BATCH_SIZE = "HINDSIGHT_API_WORKER_BATCH_SIZE"
 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"
 
 # Reflect agent settings
 ENV_REFLECT_MAX_ITERATIONS = "HINDSIGHT_API_REFLECT_MAX_ITERATIONS"
 
 # Default values
 DEFAULT_DATABASE_URL = "pg0"
+DEFAULT_DATABASE_SCHEMA = "public"
 DEFAULT_LLM_PROVIDER = "openai"
 DEFAULT_LLM_MODEL = "gpt-5-mini"
 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
+DEFAULT_LLM_MAX_BACKOFF = 60.0  # Max backoff cap in seconds for retry exponential backoff
 DEFAULT_LLM_TIMEOUT = 120.0  # seconds
 
+# Vertex AI defaults
+DEFAULT_LLM_VERTEXAI_PROJECT_ID = None  # Required for Vertex AI
+DEFAULT_LLM_VERTEXAI_REGION = "us-central1"
+DEFAULT_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY = None  # Optional, uses ADC if not set
+
 DEFAULT_EMBEDDINGS_PROVIDER = "local"
 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_OPENAI_MODEL = "text-embedding-3-small"
 DEFAULT_EMBEDDING_DIMENSION = 384
 
 DEFAULT_RERANKER_PROVIDER = "local"
 DEFAULT_RERANKER_LOCAL_MODEL = "cross-encoder/ms-marco-MiniLM-L-6-v2"
+DEFAULT_RERANKER_LOCAL_FORCE_CPU = False  # Force CPU mode for local reranker (avoids MPS/XPC issues on macOS)
 DEFAULT_RERANKER_LOCAL_MAX_CONCURRENT = 4  # Limit concurrent CPU-bound reranking to prevent thrashing
 DEFAULT_RERANKER_TEI_BATCH_SIZE = 128
 DEFAULT_RERANKER_TEI_MAX_CONCURRENT = 8
@@ -172,11 +210,11 @@ 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
 DEFAULT_RETAIN_CUSTOM_INSTRUCTIONS = None  # Custom extraction guidelines (only used when mode="custom")
-DEFAULT_RETAIN_OBSERVATIONS_ASYNC = False  # Run observation generation async (after retain completes)
 
 # Observations defaults (consolidated knowledge from facts)
 DEFAULT_ENABLE_OBSERVATIONS = True  # Observations enabled by default
 DEFAULT_CONSOLIDATION_BATCH_SIZE = 50  # Memories to load per batch (internal memory optimization)
+DEFAULT_CONSOLIDATION_MAX_TOKENS = 1024  # Max tokens for recall when finding related observations
 
 # Database migrations
 DEFAULT_RUN_MIGRATIONS_ON_STARTUP = True
@@ -192,8 +230,9 @@ DEFAULT_WORKER_ENABLED = True  # API runs worker by default (standalone mode)
 DEFAULT_WORKER_ID = None  # Will use hostname if not specified
 DEFAULT_WORKER_POLL_INTERVAL_MS = 500  # Poll database every 500ms
 DEFAULT_WORKER_MAX_RETRIES = 3  # Max retries before marking task failed
-DEFAULT_WORKER_BATCH_SIZE = 10  # Tasks to claim per poll cycle
 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
 
 # Reflect agent settings
 DEFAULT_REFLECT_MAX_ITERATIONS = 10  # Max tool call iterations before forcing response
@@ -270,6 +309,7 @@ class HindsightConfig:
 
     # Database
     database_url: str
+    database_schema: str
 
     # LLM (default, used as fallback for per-operation config)
     llm_provider: str
@@ -277,27 +317,51 @@ class HindsightConfig:
     llm_model: str
     llm_base_url: str | None
     llm_max_concurrent: int
+    llm_max_retries: int
+    llm_initial_backoff: float
+    llm_max_backoff: float
     llm_timeout: float
 
+    # Vertex AI configuration
+    llm_vertexai_project_id: str | None
+    llm_vertexai_region: str
+    llm_vertexai_service_account_key: str | None
+
     # Per-operation LLM configuration (None = use default LLM config)
     retain_llm_provider: str | None
     retain_llm_api_key: str | None
     retain_llm_model: str | None
     retain_llm_base_url: str | None
+    retain_llm_max_concurrent: int | None
+    retain_llm_max_retries: int | None
+    retain_llm_initial_backoff: float | None
+    retain_llm_max_backoff: float | None
+    retain_llm_timeout: float | None
 
     reflect_llm_provider: str | None
     reflect_llm_api_key: str | None
     reflect_llm_model: str | None
     reflect_llm_base_url: str | None
+    reflect_llm_max_concurrent: int | None
+    reflect_llm_max_retries: int | None
+    reflect_llm_initial_backoff: float | None
+    reflect_llm_max_backoff: float | None
+    reflect_llm_timeout: float | None
 
     consolidation_llm_provider: str | None
     consolidation_llm_api_key: str | None
     consolidation_llm_model: str | None
     consolidation_llm_base_url: str | None
+    consolidation_llm_max_concurrent: int | None
+    consolidation_llm_max_retries: int | None
+    consolidation_llm_initial_backoff: float | None
+    consolidation_llm_max_backoff: float | None
+    consolidation_llm_timeout: float | None
 
     # Embeddings
     embeddings_provider: str
     embeddings_local_model: str
+    embeddings_local_force_cpu: bool
     embeddings_tei_url: str | None
     embeddings_openai_base_url: str | None
     embeddings_cohere_base_url: str | None
@@ -305,6 +369,8 @@ class HindsightConfig:
     # Reranker
     reranker_provider: str
     reranker_local_model: str
+    reranker_local_force_cpu: bool
+    reranker_local_max_concurrent: int
     reranker_tei_url: str | None
     reranker_tei_batch_size: int
     reranker_tei_max_concurrent: int
@@ -331,11 +397,11 @@ class HindsightConfig:
     retain_extract_causal_links: bool
     retain_extraction_mode: str
     retain_custom_instructions: str | None
-    retain_observations_async: bool
 
     # Observations settings (consolidated knowledge from facts)
     enable_observations: bool
     consolidation_batch_size: int
+    consolidation_max_tokens: int
 
     # Optimization flags
     skip_llm_verification: bool
@@ -355,8 +421,9 @@ class HindsightConfig:
     worker_id: str | None
     worker_poll_interval_ms: int
     worker_max_retries: int
-    worker_batch_size: int
     worker_http_port: int
+    worker_max_slots: int
+    worker_consolidation_max_slots: int
 
     # Reflect agent settings
     reflect_max_iterations: int
@@ -367,35 +434,98 @@ class HindsightConfig:
         return cls(
             # Database
             database_url=os.getenv(ENV_DATABASE_URL, DEFAULT_DATABASE_URL),
+            database_schema=os.getenv(ENV_DATABASE_SCHEMA, DEFAULT_DATABASE_SCHEMA),
             # LLM
             llm_provider=os.getenv(ENV_LLM_PROVIDER, DEFAULT_LLM_PROVIDER),
             llm_api_key=os.getenv(ENV_LLM_API_KEY),
             llm_model=os.getenv(ENV_LLM_MODEL, DEFAULT_LLM_MODEL),
             llm_base_url=os.getenv(ENV_LLM_BASE_URL) or None,
             llm_max_concurrent=int(os.getenv(ENV_LLM_MAX_CONCURRENT, str(DEFAULT_LLM_MAX_CONCURRENT))),
+            llm_max_retries=int(os.getenv(ENV_LLM_MAX_RETRIES, str(DEFAULT_LLM_MAX_RETRIES))),
+            llm_initial_backoff=float(os.getenv(ENV_LLM_INITIAL_BACKOFF, str(DEFAULT_LLM_INITIAL_BACKOFF))),
+            llm_max_backoff=float(os.getenv(ENV_LLM_MAX_BACKOFF, str(DEFAULT_LLM_MAX_BACKOFF))),
             llm_timeout=float(os.getenv(ENV_LLM_TIMEOUT, str(DEFAULT_LLM_TIMEOUT))),
+            # 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),
+            llm_vertexai_service_account_key=os.getenv(ENV_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY)
+            or DEFAULT_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY,
             # 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,
             retain_llm_model=os.getenv(ENV_RETAIN_LLM_MODEL) or None,
             retain_llm_base_url=os.getenv(ENV_RETAIN_LLM_BASE_URL) or None,
+            retain_llm_max_concurrent=int(os.getenv(ENV_RETAIN_LLM_MAX_CONCURRENT))
+            if os.getenv(ENV_RETAIN_LLM_MAX_CONCURRENT)
+            else None,
+            retain_llm_max_retries=int(os.getenv(ENV_RETAIN_LLM_MAX_RETRIES))
+            if os.getenv(ENV_RETAIN_LLM_MAX_RETRIES)
+            else None,
+            retain_llm_initial_backoff=float(os.getenv(ENV_RETAIN_LLM_INITIAL_BACKOFF))
+            if os.getenv(ENV_RETAIN_LLM_INITIAL_BACKOFF)
+            else None,
+            retain_llm_max_backoff=float(os.getenv(ENV_RETAIN_LLM_MAX_BACKOFF))
+            if os.getenv(ENV_RETAIN_LLM_MAX_BACKOFF)
+            else None,
+            retain_llm_timeout=float(os.getenv(ENV_RETAIN_LLM_TIMEOUT)) if os.getenv(ENV_RETAIN_LLM_TIMEOUT) else None,
             reflect_llm_provider=os.getenv(ENV_REFLECT_LLM_PROVIDER) or None,
             reflect_llm_api_key=os.getenv(ENV_REFLECT_LLM_API_KEY) or None,
             reflect_llm_model=os.getenv(ENV_REFLECT_LLM_MODEL) or None,
             reflect_llm_base_url=os.getenv(ENV_REFLECT_LLM_BASE_URL) or None,
+            reflect_llm_max_concurrent=int(os.getenv(ENV_REFLECT_LLM_MAX_CONCURRENT))
+            if os.getenv(ENV_REFLECT_LLM_MAX_CONCURRENT)
+            else None,
+            reflect_llm_max_retries=int(os.getenv(ENV_REFLECT_LLM_MAX_RETRIES))
+            if os.getenv(ENV_REFLECT_LLM_MAX_RETRIES)
+            else None,
+            reflect_llm_initial_backoff=float(os.getenv(ENV_REFLECT_LLM_INITIAL_BACKOFF))
+            if os.getenv(ENV_REFLECT_LLM_INITIAL_BACKOFF)
+            else None,
+            reflect_llm_max_backoff=float(os.getenv(ENV_REFLECT_LLM_MAX_BACKOFF))
+            if os.getenv(ENV_REFLECT_LLM_MAX_BACKOFF)
+            else None,
+            reflect_llm_timeout=float(os.getenv(ENV_REFLECT_LLM_TIMEOUT))
+            if os.getenv(ENV_REFLECT_LLM_TIMEOUT)
+            else None,
             consolidation_llm_provider=os.getenv(ENV_CONSOLIDATION_LLM_PROVIDER) or None,
             consolidation_llm_api_key=os.getenv(ENV_CONSOLIDATION_LLM_API_KEY) or None,
             consolidation_llm_model=os.getenv(ENV_CONSOLIDATION_LLM_MODEL) or None,
             consolidation_llm_base_url=os.getenv(ENV_CONSOLIDATION_LLM_BASE_URL) or None,
+            consolidation_llm_max_concurrent=int(os.getenv(ENV_CONSOLIDATION_LLM_MAX_CONCURRENT))
+            if os.getenv(ENV_CONSOLIDATION_LLM_MAX_CONCURRENT)
+            else None,
+            consolidation_llm_max_retries=int(os.getenv(ENV_CONSOLIDATION_LLM_MAX_RETRIES))
+            if os.getenv(ENV_CONSOLIDATION_LLM_MAX_RETRIES)
+            else None,
+            consolidation_llm_initial_backoff=float(os.getenv(ENV_CONSOLIDATION_LLM_INITIAL_BACKOFF))
+            if os.getenv(ENV_CONSOLIDATION_LLM_INITIAL_BACKOFF)
+            else None,
+            consolidation_llm_max_backoff=float(os.getenv(ENV_CONSOLIDATION_LLM_MAX_BACKOFF))
+            if os.getenv(ENV_CONSOLIDATION_LLM_MAX_BACKOFF)
+            else None,
+            consolidation_llm_timeout=float(os.getenv(ENV_CONSOLIDATION_LLM_TIMEOUT))
+            if os.getenv(ENV_CONSOLIDATION_LLM_TIMEOUT)
+            else None,
             # Embeddings
             embeddings_provider=os.getenv(ENV_EMBEDDINGS_PROVIDER, DEFAULT_EMBEDDINGS_PROVIDER),
             embeddings_local_model=os.getenv(ENV_EMBEDDINGS_LOCAL_MODEL, DEFAULT_EMBEDDINGS_LOCAL_MODEL),
+            embeddings_local_force_cpu=os.getenv(
+                ENV_EMBEDDINGS_LOCAL_FORCE_CPU, str(DEFAULT_EMBEDDINGS_LOCAL_FORCE_CPU)
+            ).lower()
+            in ("true", "1"),
             embeddings_tei_url=os.getenv(ENV_EMBEDDINGS_TEI_URL),
             embeddings_openai_base_url=os.getenv(ENV_EMBEDDINGS_OPENAI_BASE_URL) or None,
             embeddings_cohere_base_url=os.getenv(ENV_EMBEDDINGS_COHERE_BASE_URL) or None,
             # Reranker
             reranker_provider=os.getenv(ENV_RERANKER_PROVIDER, DEFAULT_RERANKER_PROVIDER),
             reranker_local_model=os.getenv(ENV_RERANKER_LOCAL_MODEL, DEFAULT_RERANKER_LOCAL_MODEL),
+            reranker_local_force_cpu=os.getenv(
+                ENV_RERANKER_LOCAL_FORCE_CPU, str(DEFAULT_RERANKER_LOCAL_FORCE_CPU)
+            ).lower()
+            in ("true", "1"),
+            reranker_local_max_concurrent=int(
+                os.getenv(ENV_RERANKER_LOCAL_MAX_CONCURRENT, str(DEFAULT_RERANKER_LOCAL_MAX_CONCURRENT))
+            ),
             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(
@@ -435,15 +565,14 @@ class HindsightConfig:
                 os.getenv(ENV_RETAIN_EXTRACTION_MODE, DEFAULT_RETAIN_EXTRACTION_MODE)
             ),
             retain_custom_instructions=os.getenv(ENV_RETAIN_CUSTOM_INSTRUCTIONS) or DEFAULT_RETAIN_CUSTOM_INSTRUCTIONS,
-            retain_observations_async=os.getenv(
-                ENV_RETAIN_OBSERVATIONS_ASYNC, str(DEFAULT_RETAIN_OBSERVATIONS_ASYNC)
-            ).lower()
-            == "true",
             # Observations settings (consolidated knowledge from facts)
             enable_observations=os.getenv(ENV_ENABLE_OBSERVATIONS, str(DEFAULT_ENABLE_OBSERVATIONS)).lower() == "true",
             consolidation_batch_size=int(
                 os.getenv(ENV_CONSOLIDATION_BATCH_SIZE, str(DEFAULT_CONSOLIDATION_BATCH_SIZE))
             ),
+            consolidation_max_tokens=int(
+                os.getenv(ENV_CONSOLIDATION_MAX_TOKENS, str(DEFAULT_CONSOLIDATION_MAX_TOKENS))
+            ),
             # Database migrations
             run_migrations_on_startup=os.getenv(ENV_RUN_MIGRATIONS_ON_STARTUP, "true").lower() == "true",
             # Database connection pool
@@ -456,8 +585,11 @@ class HindsightConfig:
             worker_id=os.getenv(ENV_WORKER_ID) or DEFAULT_WORKER_ID,
             worker_poll_interval_ms=int(os.getenv(ENV_WORKER_POLL_INTERVAL_MS, str(DEFAULT_WORKER_POLL_INTERVAL_MS))),
             worker_max_retries=int(os.getenv(ENV_WORKER_MAX_RETRIES, str(DEFAULT_WORKER_MAX_RETRIES))),
-            worker_batch_size=int(os.getenv(ENV_WORKER_BATCH_SIZE, str(DEFAULT_WORKER_BATCH_SIZE))),
             worker_http_port=int(os.getenv(ENV_WORKER_HTTP_PORT, str(DEFAULT_WORKER_HTTP_PORT))),
+            worker_max_slots=int(os.getenv(ENV_WORKER_MAX_SLOTS, str(DEFAULT_WORKER_MAX_SLOTS))),
+            worker_consolidation_max_slots=int(
+                os.getenv(ENV_WORKER_CONSOLIDATION_MAX_SLOTS, str(DEFAULT_WORKER_CONSOLIDATION_MAX_SLOTS))
+            ),
             # Reflect agent settings
             reflect_max_iterations=int(os.getenv(ENV_REFLECT_MAX_ITERATIONS, str(DEFAULT_REFLECT_MAX_ITERATIONS))),
         )
@@ -515,7 +647,7 @@ class HindsightConfig:
 
     def log_config(self) -> None:
         """Log the current configuration (without sensitive values)."""
-        logger.info(f"Database: {self.database_url}")
+        logger.info(f"Database: {self.database_url} (schema: {self.database_schema})")
         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/hindsight_api/daemon.py

@@ -52,7 +52,10 @@ class IdleTimeoutMiddleware:
                 logger.info(f"Idle timeout reached ({self.idle_timeout}s), shutting down daemon")
                 # Give a moment for any in-flight requests
                 await asyncio.sleep(1)
-                os._exit(0)
+                # Send SIGTERM to ourselves to trigger graceful shutdown
+                import signal
+
+                os.kill(os.getpid(), signal.SIGTERM)
 
 
 class DaemonLock:

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

@@ -144,10 +144,14 @@ async def run_consolidation_job(
     }
 
     batch_num = 0
+    last_progress_timings = {}  # Track timings at last progress log
     while True:
         batch_num += 1
         batch_start = time.time()
 
+        # Snapshot timings at batch start for per-batch calculation
+        batch_start_timings = perf.timings.copy()
+
         # Fetch next batch of unconsolidated memories
         async with pool.acquire() as conn:
             t0 = time.time()
@@ -217,19 +221,44 @@ async def run_consolidation_job(
             elif action == "skipped":
                 stats["skipped"] += 1
 
-            # Log progress periodically
+            # Log progress periodically with timing breakdown
             if stats["memories_processed"] % 10 == 0:
+                # Calculate timing deltas since last progress log
+                timing_parts = []
+                for key in ["recall", "llm", "embedding", "db_write"]:
+                    if key in perf.timings:
+                        delta = perf.timings[key] - last_progress_timings.get(key, 0)
+                        timing_parts.append(f"{key}={delta:.2f}s")
+
+                timing_str = f" | {', '.join(timing_parts)}" if timing_parts else ""
                 logger.info(
                     f"[CONSOLIDATION] bank={bank_id} progress: "
-                    f"{stats['memories_processed']}/{total_count} memories processed"
+                    f"{stats['memories_processed']}/{total_count} memories processed{timing_str}"
                 )
 
+                # Update last progress snapshot
+                last_progress_timings = perf.timings.copy()
+
         batch_time = time.time() - batch_start
         perf.log(
             f"[2] Batch {batch_num}: {len(memories)} memories in {batch_time:.3f}s "
             f"(avg {batch_time / len(memories):.3f}s/memory)"
         )
 
+        # Log timing breakdown after each batch (delta from batch start)
+        timing_parts = []
+        for key in ["recall", "llm", "embedding", "db_write"]:
+            if key in perf.timings:
+                delta = perf.timings[key] - batch_start_timings.get(key, 0)
+                timing_parts.append(f"{key}={delta:.3f}s")
+
+        if timing_parts:
+            avg_per_memory = batch_time / len(memories) if memories else 0
+            logger.info(
+                f"[CONSOLIDATION] bank={bank_id} batch {batch_num}/{len(memories)} memories: "
+                f"{', '.join(timing_parts)} | avg={avg_per_memory:.3f}s/memory"
+            )
+
     # Build summary
     perf.log(
         f"[3] Results: {stats['memories_processed']} memories -> "
@@ -639,28 +668,27 @@ async def _find_related_observations(
     request_context: "RequestContext",
 ) -> list[dict[str, Any]]:
     """
-    Find observations related to the given query using the full recall system.
+    Find observations related to the given query using optimized recall.
 
     IMPORTANT: We do NOT filter by tags here. Consolidation needs to see ALL
     potentially related observations regardless of scope, so the LLM can
     decide on tag routing (same scope update vs cross-scope create).
 
-    This leverages:
-    - Semantic search (embedding similarity)
-    - BM25 text search (keyword matching)
-    - Entity-based retrieval (shared entities)
-    - Graph traversal (connected via entity links)
+    Uses max_tokens to naturally limit observations (no artificial count limit).
+    Includes source memories with dates for LLM context.
 
     Returns:
-        List of related observations with their tags for LLM tag routing
+        List of related observations with their tags, source memories, and dates
     """
-    # Use recall to find related observations
-    # NO tags parameter - we want ALL observations regardless of scope
-    # Use low max_tokens since we only need observations, not memories
+    # Use recall to find related observations with token budget
+    # max_tokens naturally limits how many observations are returned
+    from ...config import get_config
+
+    config = get_config()
     recall_result = await memory_engine.recall_async(
         bank_id=bank_id,
         query=query,
-        max_tokens=5000,  # Token budget for observations
+        max_tokens=config.consolidation_max_tokens,  # Token budget for observations (configurable)
         fact_type=["observation"],  # Only retrieve observations
         request_context=request_context,
         _quiet=True,  # Suppress logging
@@ -668,43 +696,82 @@ async def _find_related_observations(
     )
 
     # If no observations returned, return empty list
-    # When fact_type=["observation"], results come back in `results` field
     if not recall_result.results:
         return []
 
-    # Trust recall's relevance filtering - fetch full data for each observation
+    # Batch fetch all observations in a single query (no artificial limit)
+    observation_ids = [uuid.UUID(obs.id) for obs in recall_result.results]
+
+    rows = await conn.fetch(
+        f"""
+        SELECT id, text, proof_count, history, tags, source_memory_ids, created_at, updated_at,
+               occurred_start, occurred_end, mentioned_at
+        FROM {fq_table("memory_units")}
+        WHERE id = ANY($1) AND bank_id = $2 AND fact_type = 'observation'
+        """,
+        observation_ids,
+        bank_id,
+    )
+
+    # Build results list preserving recall order
+    id_to_row = {row["id"]: row for row in rows}
     results = []
+
     for obs in recall_result.results:
-        # Fetch full observation data from DB to get history, source_memory_ids, tags
-        row = await conn.fetchrow(
-            f"""
-            SELECT id, text, proof_count, history, tags, source_memory_ids, created_at, updated_at
-            FROM {fq_table("memory_units")}
-            WHERE id = $1 AND bank_id = $2 AND fact_type = 'observation'
-            """,
-            uuid.UUID(obs.id),
-            bank_id,
-        )
+        obs_id = uuid.UUID(obs.id)
+        if obs_id not in id_to_row:
+            continue
 
-        if row:
-            history = row["history"]
-            if isinstance(history, str):
-                history = json.loads(history)
-            elif history is None:
-                history = []
+        row = id_to_row[obs_id]
+        history = row["history"]
+        if isinstance(history, str):
+            history = json.loads(history)
+        elif history is None:
+            history = []
 
-            results.append(
-                {
-                    "id": row["id"],
-                    "text": row["text"],
-                    "proof_count": row["proof_count"] or 1,
-                    "history": history,
-                    "tags": row["tags"] or [],  # Include tags for LLM tag routing
-                    "source_memory_ids": row["source_memory_ids"] or [],
-                    "similarity": 1.0,  # Retrieved via recall so assumed relevant
-                }
+        # Fetch source memories to include their text and dates
+        source_memory_ids = row["source_memory_ids"] or []
+        source_memories = []
+
+        if source_memory_ids:
+            source_rows = await conn.fetch(
+                f"""
+                SELECT text, occurred_start, occurred_end, mentioned_at, event_date
+                FROM {fq_table("memory_units")}
+                WHERE id = ANY($1) AND bank_id = $2
+                ORDER BY created_at ASC
+                LIMIT 5
+                """,
+                source_memory_ids[:5],  # Limit to first 5 source memories for token efficiency
+                bank_id,
             )
 
+            for src_row in source_rows:
+                source_memories.append(
+                    {
+                        "text": src_row["text"],
+                        "occurred_start": src_row["occurred_start"],
+                        "occurred_end": src_row["occurred_end"],
+                        "mentioned_at": src_row["mentioned_at"],
+                        "event_date": src_row["event_date"],
+                    }
+                )
+
+        results.append(
+            {
+                "id": row["id"],
+                "text": row["text"],
+                "proof_count": row["proof_count"] or 1,
+                "tags": row["tags"] or [],
+                "source_memories": source_memories,
+                "occurred_start": row["occurred_start"],
+                "occurred_end": row["occurred_end"],
+                "mentioned_at": row["mentioned_at"],
+                "created_at": row["created_at"],
+                "updated_at": row["updated_at"],
+            }
+        )
+
     return results
 
 
@@ -732,14 +799,43 @@ async def _consolidate_with_llm(
         - {"action": "create", "text": "...", "reason": "..."}
         - [] if fact is purely ephemeral (no durable knowledge)
     """
-    # Format observations WITH their tags (or "None" if empty)
+    # Format observations as JSON with source memories and dates
     if observations:
-        observations_text = "\n".join(
-            f'- ID: {obs["id"]}, Tags: {json.dumps(obs["tags"])}, Text: "{obs["text"]}" (proof_count: {obs["proof_count"]})'
-            for obs in observations
-        )
+        obs_list = []
+        for obs in observations:
+            obs_data = {
+                "id": str(obs["id"]),
+                "text": obs["text"],
+                "proof_count": obs["proof_count"],
+                "tags": obs["tags"],
+                "created_at": obs["created_at"].isoformat() if obs.get("created_at") else None,
+                "updated_at": obs["updated_at"].isoformat() if obs.get("updated_at") else None,
+            }
+
+            # Include temporal info if available
+            if obs.get("occurred_start"):
+                obs_data["occurred_start"] = obs["occurred_start"].isoformat()
+            if obs.get("occurred_end"):
+                obs_data["occurred_end"] = obs["occurred_end"].isoformat()
+            if obs.get("mentioned_at"):
+                obs_data["mentioned_at"] = obs["mentioned_at"].isoformat()
+
+            # Include source memories (up to 3 for brevity)
+            if obs.get("source_memories"):
+                obs_data["source_memories"] = [
+                    {
+                        "text": sm["text"],
+                        "event_date": sm["event_date"].isoformat() if sm.get("event_date") else None,
+                        "occurred_start": sm["occurred_start"].isoformat() if sm.get("occurred_start") else None,
+                    }
+                    for sm in obs["source_memories"][:3]  # Limit to 3 for token efficiency
+                ]
+
+            obs_list.append(obs_data)
+
+        observations_text = json.dumps(obs_list, indent=2)
     else:
-        observations_text = "None (this is a new topic - create if fact contains durable knowledge)"
+        observations_text = "[]"
 
     # Only include mission section if mission is set and not the default
     mission_section = ""
@@ -769,7 +865,14 @@ Focus on DURABLE knowledge that serves this mission, not ephemeral state.
         )
         # Parse JSON response - should be an array
         if isinstance(result, str):
-            result = json.loads(result)
+            # Strip markdown code fences (some models wrap JSON in ```json ... ```)
+            clean = result.strip()
+            if clean.startswith("```"):
+                clean = clean.split("\n", 1)[1] if "\n" in clean else clean[3:]
+                if clean.endswith("```"):
+                    clean = clean[:-3]
+                clean = clean.strip()
+            result = json.loads(clean)
         # Ensure result is a list
         if isinstance(result, list):
             return result

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

@@ -47,23 +47,31 @@ CONSOLIDATION_USER_PROMPT = """Analyze this new fact and consolidate into knowle
 {mission_section}
 NEW FACT: {fact_text}
 
-EXISTING OBSERVATIONS:
+EXISTING OBSERVATIONS (JSON array with source memories and dates):
 {observations_text}
 
-Instructions:
-1. First, extract the DURABLE KNOWLEDGE from the fact (not ephemeral state like "user is at X")
-2. Then compare with existing observations:
-   - If an observation covers the same topic: UPDATE it with the new knowledge
-   - If no observation covers the topic: CREATE a new one
+Each observation includes:
+- id: unique identifier for updating
+- text: the observation content
+- proof_count: number of supporting memories
+- tags: visibility scope (handled automatically)
+- created_at/updated_at: when observation was created/modified
+- occurred_start/occurred_end: temporal range of source facts
+- source_memories: array of supporting facts with their text and dates
 
-Output JSON array of actions (ALWAYS an array, even for single action):
+Instructions:
+1. Extract DURABLE KNOWLEDGE from the new fact (not ephemeral state)
+2. Review source_memories in existing observations to understand evidence
+3. Check dates to detect contradictions or updates
+4. Compare with observations:
+   - Same topic → UPDATE with learning_id
+   - New topic → CREATE new observation
+   - Purely ephemeral → return []
+
+Output JSON array of actions:
 [
-  {{"action": "update", "learning_id": "uuid", "text": "updated durable knowledge", "reason": "..."}},
+  {{"action": "update", "learning_id": "uuid-from-observations", "text": "updated knowledge", "reason": "..."}},
   {{"action": "create", "text": "new durable knowledge", "reason": "..."}}
 ]
 
-If NO consolidation is needed (fact is purely ephemeral with no durable knowledge):
-[]
-
-If no observations exist and fact contains durable knowledge:
-[{{"action": "create", "text": "durable knowledge text", "reason": "new topic"}}]"""
+Return [] if fact contains no durable knowledge."""

hindsight-api/hindsight_api/engine/cross_encoder.py

@@ -20,6 +20,7 @@ from ..config import (
     DEFAULT_RERANKER_FLASHRANK_CACHE_DIR,
     DEFAULT_RERANKER_FLASHRANK_MODEL,
     DEFAULT_RERANKER_LITELLM_MODEL,
+    DEFAULT_RERANKER_LOCAL_FORCE_CPU,
     DEFAULT_RERANKER_LOCAL_MAX_CONCURRENT,
     DEFAULT_RERANKER_LOCAL_MODEL,
     DEFAULT_RERANKER_PROVIDER,
@@ -33,6 +34,7 @@ from ..config import (
     ENV_RERANKER_FLASHRANK_CACHE_DIR,
     ENV_RERANKER_FLASHRANK_MODEL,
     ENV_RERANKER_LITELLM_MODEL,
+    ENV_RERANKER_LOCAL_FORCE_CPU,
     ENV_RERANKER_LOCAL_MAX_CONCURRENT,
     ENV_RERANKER_LOCAL_MODEL,
     ENV_RERANKER_PROVIDER,
@@ -99,7 +101,7 @@ class LocalSTCrossEncoder(CrossEncoderModel):
     _executor: ThreadPoolExecutor | None = None
     _max_concurrent: int = 4  # Limit concurrent CPU-bound reranking calls
 
-    def __init__(self, model_name: str | None = None, max_concurrent: int = 4):
+    def __init__(self, model_name: str | None = None, max_concurrent: int = 4, force_cpu: bool = False):
         """
         Initialize local SentenceTransformers cross-encoder.
 
@@ -108,8 +110,11 @@ class LocalSTCrossEncoder(CrossEncoderModel):
                        Default: cross-encoder/ms-marco-MiniLM-L-6-v2
             max_concurrent: Maximum concurrent reranking calls (default: 2).
                            Higher values may cause CPU thrashing under load.
+            force_cpu: Force CPU mode (avoids MPS/XPC issues on macOS in daemon mode).
+                      Default: False
         """
         self.model_name = model_name or DEFAULT_RERANKER_LOCAL_MODEL
+        self.force_cpu = force_cpu
         self._model = None
         LocalSTCrossEncoder._max_concurrent = max_concurrent
 
@@ -139,13 +144,23 @@ class LocalSTCrossEncoder(CrossEncoderModel):
         # after loading, which conflicts with accelerate's device_map handling.
         import torch
 
-        # Check for GPU (CUDA) or Apple Silicon (MPS)
-        has_gpu = torch.cuda.is_available() or (hasattr(torch.backends, "mps") and torch.backends.mps.is_available())
-
-        if has_gpu:
-            device = None  # Let sentence-transformers auto-detect GPU/MPS
-        else:
+        # Force CPU mode if configured (used in daemon mode to avoid MPS/XPC issues on macOS)
+        if self.force_cpu:
             device = "cpu"
+            logger.info("Reranker: forcing CPU mode (HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU=1)")
+        else:
+            # Check for GPU (CUDA) or Apple Silicon (MPS)
+            # Wrap in try-except to gracefully handle any device detection issues
+            # (e.g., in CI environments or when PyTorch is built without GPU support)
+            device = "cpu"  # Default to CPU
+            try:
+                has_gpu = torch.cuda.is_available() or (
+                    hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
+                )
+                if has_gpu:
+                    device = None  # Let sentence-transformers auto-detect GPU/MPS
+            except Exception as e:
+                logger.warning(f"Failed to detect GPU/MPS, falling back to CPU: {e}")
 
         self._model = CrossEncoder(
             self.model_name,
@@ -163,6 +178,11 @@ class LocalSTCrossEncoder(CrossEncoderModel):
         else:
             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)
+        return scores.tolist() if hasattr(scores, "tolist") else list(scores)
+
     async def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
         """
         Score query-document pairs for relevance.
@@ -180,11 +200,11 @@ class LocalSTCrossEncoder(CrossEncoderModel):
 
         # Use dedicated executor - limited workers naturally limits concurrency
         loop = asyncio.get_event_loop()
-        scores = await loop.run_in_executor(
+        return await loop.run_in_executor(
             LocalSTCrossEncoder._executor,
-            lambda: self._model.predict(pairs, show_progress_bar=False),
+            self._predict_sync,
+            pairs,
         )
-        return scores.tolist() if hasattr(scores, "tolist") else list(scores)
 
 
 class RemoteTEICrossEncoder(CrossEncoderModel):
@@ -594,7 +614,7 @@ class FlashRankCrossEncoder(CrossEncoderModel):
             return
 
         try:
-            from flashrank import Ranker  # type: ignore[import-untyped]
+            from flashrank import Ranker
         except ImportError:
             raise ImportError("flashrank is required for FlashRankCrossEncoder. Install it with: pip install flashrank")
 
@@ -621,7 +641,7 @@ class FlashRankCrossEncoder(CrossEncoderModel):
 
     def _predict_sync(self, pairs: list[tuple[str, str]]) -> list[float]:
         """Synchronous predict - processes each query group."""
-        from flashrank import RerankRequest  # type: ignore[import-untyped]
+        from flashrank import RerankRequest
 
         if not pairs:
             return []
@@ -783,29 +803,33 @@ class LiteLLMCrossEncoder(CrossEncoderModel):
 
 def create_cross_encoder_from_env() -> CrossEncoderModel:
     """
-    Create a CrossEncoderModel instance based on environment variables.
+    Create a CrossEncoderModel instance based on configuration.
 
-    See hindsight_api.config for environment variable names and defaults.
+    Reads configuration via get_config() to ensure consistency across the codebase.
 
     Returns:
         Configured CrossEncoderModel instance
     """
-    provider = os.environ.get(ENV_RERANKER_PROVIDER, DEFAULT_RERANKER_PROVIDER).lower()
+    from ..config import get_config
+
+    config = get_config()
+    provider = config.reranker_provider.lower()
 
     if provider == "tei":
-        url = os.environ.get(ENV_RERANKER_TEI_URL)
+        url = config.reranker_tei_url
         if not url:
             raise ValueError(f"{ENV_RERANKER_TEI_URL} is required when {ENV_RERANKER_PROVIDER} is 'tei'")
-        batch_size = int(os.environ.get(ENV_RERANKER_TEI_BATCH_SIZE, str(DEFAULT_RERANKER_TEI_BATCH_SIZE)))
-        max_concurrent = int(os.environ.get(ENV_RERANKER_TEI_MAX_CONCURRENT, str(DEFAULT_RERANKER_TEI_MAX_CONCURRENT)))
-        return RemoteTEICrossEncoder(base_url=url, batch_size=batch_size, max_concurrent=max_concurrent)
+        return RemoteTEICrossEncoder(
+            base_url=url,
+            batch_size=config.reranker_tei_batch_size,
+            max_concurrent=config.reranker_tei_max_concurrent,
+        )
     elif provider == "local":
-        model = os.environ.get(ENV_RERANKER_LOCAL_MODEL)
-        model_name = model or DEFAULT_RERANKER_LOCAL_MODEL
-        max_concurrent = int(
-            os.environ.get(ENV_RERANKER_LOCAL_MAX_CONCURRENT, str(DEFAULT_RERANKER_LOCAL_MAX_CONCURRENT))
+        return LocalSTCrossEncoder(
+            model_name=config.reranker_local_model,
+            max_concurrent=config.reranker_local_max_concurrent,
+            force_cpu=config.reranker_local_force_cpu,
         )
-        return LocalSTCrossEncoder(model_name=model_name, max_concurrent=max_concurrent)
     elif provider == "cohere":
         api_key = os.environ.get(ENV_COHERE_API_KEY)
         if not api_key:

hindsight-api/hindsight_api/engine/embeddings.py

@@ -18,6 +18,7 @@ import httpx
 from ..config import (
     DEFAULT_EMBEDDINGS_COHERE_MODEL,
     DEFAULT_EMBEDDINGS_LITELLM_MODEL,
+    DEFAULT_EMBEDDINGS_LOCAL_FORCE_CPU,
     DEFAULT_EMBEDDINGS_LOCAL_MODEL,
     DEFAULT_EMBEDDINGS_OPENAI_MODEL,
     DEFAULT_EMBEDDINGS_PROVIDER,
@@ -26,6 +27,7 @@ from ..config import (
     ENV_EMBEDDINGS_COHERE_BASE_URL,
     ENV_EMBEDDINGS_COHERE_MODEL,
     ENV_EMBEDDINGS_LITELLM_MODEL,
+    ENV_EMBEDDINGS_LOCAL_FORCE_CPU,
     ENV_EMBEDDINGS_LOCAL_MODEL,
     ENV_EMBEDDINGS_OPENAI_API_KEY,
     ENV_EMBEDDINGS_OPENAI_BASE_URL,
@@ -92,15 +94,18 @@ class LocalSTEmbeddings(Embeddings):
     The embedding dimension is auto-detected from the model.
     """
 
-    def __init__(self, model_name: str | None = None):
+    def __init__(self, model_name: str | None = None, force_cpu: bool = False):
         """
         Initialize local SentenceTransformers embeddings.
 
         Args:
             model_name: Name of the SentenceTransformer model to use.
                        Default: BAAI/bge-small-en-v1.5
+            force_cpu: Force CPU mode (avoids MPS/XPC issues on macOS in daemon mode).
+                      Default: False
         """
         self.model_name = model_name or DEFAULT_EMBEDDINGS_LOCAL_MODEL
+        self.force_cpu = force_cpu
         self._model = None
         self._dimension: int | None = None
 
@@ -134,13 +139,23 @@ class LocalSTEmbeddings(Embeddings):
         # which can cause issues when accelerate is installed but no GPU is available.
         import torch
 
-        # Check for GPU (CUDA) or Apple Silicon (MPS)
-        has_gpu = torch.cuda.is_available() or (hasattr(torch.backends, "mps") and torch.backends.mps.is_available())
-
-        if has_gpu:
-            device = None  # Let sentence-transformers auto-detect GPU/MPS
-        else:
+        # Force CPU mode if configured (used in daemon mode to avoid MPS/XPC issues on macOS)
+        if self.force_cpu:
             device = "cpu"
+            logger.info("Embeddings: forcing CPU mode")
+        else:
+            # Check for GPU (CUDA) or Apple Silicon (MPS)
+            # Wrap in try-except to gracefully handle any device detection issues
+            # (e.g., in CI environments or when PyTorch is built without GPU support)
+            device = "cpu"  # Default to CPU
+            try:
+                has_gpu = torch.cuda.is_available() or (
+                    hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
+                )
+                if has_gpu:
+                    device = None  # Let sentence-transformers auto-detect GPU/MPS
+            except Exception as e:
+                logger.warning(f"Failed to detect GPU/MPS, falling back to CPU: {e}")
 
         self._model = SentenceTransformer(
             self.model_name,
@@ -163,6 +178,7 @@ class LocalSTEmbeddings(Embeddings):
         """
         if self._model is None:
             raise RuntimeError("Embeddings not initialized. Call initialize() first.")
+
         embeddings = self._model.encode(texts, convert_to_numpy=True, show_progress_bar=False)
         return [emb.tolist() for emb in embeddings]
 
@@ -529,7 +545,7 @@ class CohereEmbeddings(Embeddings):
                 model=self.model,
                 input_type=self.input_type,
             )
-            if response.embeddings:
+            if response.embeddings and isinstance(response.embeddings, list):
                 self._dimension = len(response.embeddings[0])
 
         logger.info(f"Embeddings: Cohere provider initialized (model: {self.model}, dim: {self._dimension})")
@@ -686,24 +702,28 @@ class LiteLLMEmbeddings(Embeddings):
 
 def create_embeddings_from_env() -> Embeddings:
     """
-    Create an Embeddings instance based on environment variables.
+    Create an Embeddings instance based on configuration.
 
-    See hindsight_api.config for environment variable names and defaults.
+    Reads configuration via get_config() to ensure consistency across the codebase.
 
     Returns:
         Configured Embeddings instance
     """
-    provider = os.environ.get(ENV_EMBEDDINGS_PROVIDER, DEFAULT_EMBEDDINGS_PROVIDER).lower()
+    from ..config import get_config
+
+    config = get_config()
+    provider = config.embeddings_provider.lower()
 
     if provider == "tei":
-        url = os.environ.get(ENV_EMBEDDINGS_TEI_URL)
+        url = config.embeddings_tei_url
         if not url:
             raise ValueError(f"{ENV_EMBEDDINGS_TEI_URL} is required when {ENV_EMBEDDINGS_PROVIDER} is 'tei'")
         return RemoteTEIEmbeddings(base_url=url)
     elif provider == "local":
-        model = os.environ.get(ENV_EMBEDDINGS_LOCAL_MODEL)
-        model_name = model or DEFAULT_EMBEDDINGS_LOCAL_MODEL
-        return LocalSTEmbeddings(model_name=model_name)
+        return LocalSTEmbeddings(
+            model_name=config.embeddings_local_model,
+            force_cpu=config.embeddings_local_force_cpu,
+        )
     elif provider == "openai":
         # Use dedicated embeddings API key, or fall back to LLM API key
         api_key = os.environ.get(ENV_EMBEDDINGS_OPENAI_API_KEY) or os.environ.get(ENV_LLM_API_KEY)

hindsight-api/hindsight_api/engine/interface.py

@@ -442,49 +442,6 @@ class MemoryEngineInterface(ABC):
         """
         ...
 
-    @abstractmethod
-    async def get_entity_observations(
-        self,
-        bank_id: str,
-        entity_id: str,
-        *,
-        limit: int = 10,
-        request_context: "RequestContext",
-    ) -> list[Any]:
-        """
-        Get observations for an entity.
-
-        Args:
-            bank_id: The memory bank ID.
-            entity_id: The entity ID.
-            limit: Maximum observations.
-            request_context: Request context for authentication.
-
-        Returns:
-            List of EntityObservation objects.
-        """
-        ...
-
-    @abstractmethod
-    async def regenerate_entity_observations(
-        self,
-        bank_id: str,
-        entity_id: str,
-        entity_name: str,
-        *,
-        request_context: "RequestContext",
-    ) -> None:
-        """
-        Regenerate observations for an entity.
-
-        Args:
-            bank_id: The memory bank ID.
-            entity_id: The entity ID.
-            entity_name: The entity's canonical name.
-            request_context: Request context for authentication.
-        """
-        ...
-
     # =========================================================================
     # Statistics & Operations
     # =========================================================================

hindsight-api/hindsight_api/engine/llm_wrapper.py

@@ -16,6 +16,15 @@ from google.genai import errors as genai_errors
 from google.genai import types as genai_types
 from openai import APIConnectionError, APIStatusError, AsyncOpenAI, LengthFinishReasonError
 
+# Vertex AI imports (conditional)
+try:
+    import google.auth
+    from google.oauth2 import service_account
+
+    VERTEXAI_AVAILABLE = True
+except ImportError:
+    VERTEXAI_AVAILABLE = False
+
 from ..config import (
     DEFAULT_LLM_MAX_CONCURRENT,
     DEFAULT_LLM_TIMEOUT,
@@ -88,7 +97,7 @@ class LLMProvider:
         self.groq_service_tier = groq_service_tier or os.getenv(ENV_LLM_GROQ_SERVICE_TIER, "auto")
 
         # Validate provider
-        valid_providers = ["openai", "groq", "ollama", "gemini", "anthropic", "lmstudio", "mock"]
+        valid_providers = ["openai", "groq", "ollama", "gemini", "anthropic", "lmstudio", "vertexai", "mock"]
         if self.provider not in valid_providers:
             raise ValueError(f"Invalid LLM provider: {self.provider}. Must be one of: {', '.join(valid_providers)}")
 
@@ -105,8 +114,51 @@ class LLMProvider:
             elif self.provider == "lmstudio":
                 self.base_url = "http://localhost:1234/v1"
 
-        # Validate API key (not needed for ollama, lmstudio, or mock)
-        if self.provider not in ("ollama", "lmstudio", "mock") and not self.api_key:
+        # Vertex AI config — stored for client creation below
+        self._vertexai_project_id: str | None = None
+        self._vertexai_region: str | None = None
+        self._vertexai_credentials: Any = None
+
+        if self.provider == "vertexai":
+            from ..config import get_config
+
+            config = get_config()
+
+            self._vertexai_project_id = config.llm_vertexai_project_id
+            if not self._vertexai_project_id:
+                raise ValueError(
+                    "HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID is required for Vertex AI provider. "
+                    "Set it to your GCP project ID."
+                )
+
+            self._vertexai_region = config.llm_vertexai_region or "us-central1"
+            service_account_key = config.llm_vertexai_service_account_key
+
+            # Load explicit service account credentials if provided
+            if service_account_key:
+                if not VERTEXAI_AVAILABLE:
+                    raise ValueError(
+                        "Vertex AI service account auth requires 'google-auth' package. "
+                        "Install with: pip install google-auth"
+                    )
+                self._vertexai_credentials = service_account.Credentials.from_service_account_file(
+                    service_account_key,
+                    scopes=["https://www.googleapis.com/auth/cloud-platform"],
+                )
+                logger.info(f"Vertex AI: Using service account key: {service_account_key}")
+
+            # Strip google/ prefix from model name — native SDK uses bare names
+            # e.g. "google/gemini-2.0-flash-lite-001" -> "gemini-2.0-flash-lite-001"
+            if self.model.startswith("google/"):
+                self.model = self.model[len("google/") :]
+
+            logger.info(
+                f"Vertex AI: project={self._vertexai_project_id}, region={self._vertexai_region}, "
+                f"model={self.model}, auth={'service_account' if service_account_key else 'ADC'}"
+            )
+
+        # Validate API key (not needed for ollama, lmstudio, vertexai, or mock)
+        if self.provider not in ("ollama", "lmstudio", "vertexai", "mock") and not self.api_key:
             raise ValueError(f"API key not found for {self.provider}")
 
         # Get timeout config (set HINDSIGHT_API_LLM_TIMEOUT for local LLMs that need longer timeouts)
@@ -132,6 +184,17 @@ class LLMProvider:
             if self.timeout:
                 anthropic_kwargs["timeout"] = self.timeout
             self._anthropic_client = AsyncAnthropic(**anthropic_kwargs)
+        elif self.provider == "vertexai":
+            # Native genai SDK with Vertex AI — handles ADC automatically,
+            # or uses explicit service account credentials if provided
+            client_kwargs = {
+                "vertexai": True,
+                "project": self._vertexai_project_id,
+                "location": self._vertexai_region,
+            }
+            if self._vertexai_credentials is not None:
+                client_kwargs["credentials"] = self._vertexai_credentials
+            self._gemini_client = genai.Client(**client_kwargs)
         elif self.provider in ("ollama", "lmstudio"):
             # Use dummy key if not provided for local
             api_key = self.api_key or "local"
@@ -223,8 +286,8 @@ class LLMProvider:
                     return_usage,
                 )
 
-            # Handle Gemini provider separately
-            if self.provider == "gemini":
+            # Handle Gemini and Vertex AI providers (both use native genai SDK)
+            if self.provider in ("gemini", "vertexai"):
                 return await self._call_gemini(
                     messages,
                     response_format,
@@ -342,11 +405,13 @@ class LLMProvider:
                         schema_msg = f"\n\nYou must respond with valid JSON matching this schema:\n{json.dumps(schema, indent=2)}"
 
                         if call_params["messages"] and call_params["messages"][0].get("role") == "system":
-                            call_params["messages"][0]["content"] += schema_msg
+                            first_msg = call_params["messages"][0]
+                            if isinstance(first_msg, dict) and isinstance(first_msg.get("content"), str):
+                                first_msg["content"] += schema_msg
                         elif call_params["messages"]:
-                            call_params["messages"][0]["content"] = (
-                                schema_msg + "\n\n" + call_params["messages"][0]["content"]
-                            )
+                            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
                         # We rely on the schema in the system message instead
@@ -586,8 +651,8 @@ class LLMProvider:
                     messages, tools, max_completion_tokens, max_retries, initial_backoff, max_backoff, start_time, scope
                 )
 
-            # Handle Gemini (convert to Gemini tool format)
-            if self.provider == "gemini":
+            # Handle Gemini and Vertex AI (convert to Gemini tool format)
+            if self.provider in ("gemini", "vertexai"):
                 return await self._call_with_tools_gemini(
                     messages, tools, max_retries, initial_backoff, max_backoff, start_time, scope
                 )
@@ -917,18 +982,20 @@ class LLMProvider:
                 tool_calls: list[LLMToolCall] = []
 
                 if response.candidates and response.candidates[0].content:
-                    for part in response.candidates[0].content.parts:
-                        if hasattr(part, "text") and part.text:
-                            content = part.text
-                        if hasattr(part, "function_call") and part.function_call:
-                            fc = part.function_call
-                            tool_calls.append(
-                                LLMToolCall(
-                                    id=f"gemini_{len(tool_calls)}",
-                                    name=fc.name,
-                                    arguments=dict(fc.args) if fc.args else {},
+                    parts = response.candidates[0].content.parts
+                    if parts:
+                        for part in parts:
+                            if hasattr(part, "text") and part.text:
+                                content = part.text
+                            if hasattr(part, "function_call") and part.function_call:
+                                fc = part.function_call
+                                tool_calls.append(
+                                    LLMToolCall(
+                                        id=f"gemini_{len(tool_calls)}",
+                                        name=fc.name,
+                                        arguments=dict(fc.args) if fc.args else {},
+                                    )
                                 )
-                            )
 
                 finish_reason = "tool_calls" if tool_calls else "stop"
 
@@ -1504,6 +1571,10 @@ class LLMProvider:
         """Clear the recorded mock calls."""
         self._mock_calls = []
 
+    async def cleanup(self) -> None:
+        """Clean up resources."""
+        pass
+
     @classmethod
     def for_memory(cls) -> "LLMProvider":
         """Create provider for memory operations from environment variables."""

hindsight-api/hindsight_api/engine/memory_engine.py

@@ -23,12 +23,17 @@ from ..metrics import get_metrics_collector
 from .db_budget import budgeted_operation
 
 # Context variable for current schema (async-safe, per-task isolation)
-_current_schema: contextvars.ContextVar[str] = contextvars.ContextVar("current_schema", default="public")
+# Note: default is None, actual default comes from config via get_current_schema()
+_current_schema: contextvars.ContextVar[str | None] = contextvars.ContextVar("current_schema", default=None)
 
 
 def get_current_schema() -> str:
-    """Get the current schema from context (default: 'public')."""
-    return _current_schema.get()
+    """Get the current schema from context (falls back to config default)."""
+    schema = _current_schema.get()
+    if schema is None:
+        # Fall back to configured default schema
+        return get_config().database_schema
+    return schema
 
 
 def fq_table(table_name: str) -> str:
@@ -499,12 +504,11 @@ class MemoryEngine(MemoryEngineInterface):
         if request_context is None:
             raise AuthenticationError("RequestContext is required when tenant extension is configured")
 
-        # For internal/background operations (e.g., worker tasks), skip extension authentication
-        # if the schema has already been set by execute_task via the _schema field.
+        # For internal/background operations (e.g., worker tasks), skip extension authentication.
+        # The task was already authenticated at submission time, and execute_task sets _current_schema
+        # from the task's _schema field. For public schema tasks, _current_schema keeps its default "public".
         if request_context.internal:
-            current = _current_schema.get()
-            if current and current != "public":
-                return current
+            return _current_schema.get()
 
         # Let AuthenticationError propagate - HTTP layer will convert to 401
         tenant_context = await self._tenant_extension.authenticate(request_context)
@@ -784,7 +788,7 @@ class MemoryEngine(MemoryEngineInterface):
                 kwargs = {"name": self._pg0_instance_name}
                 if self._pg0_port is not None:
                     kwargs["port"] = self._pg0_port
-                pg0 = EmbeddedPostgres(**kwargs)  # type: ignore[invalid-argument-type] - dict kwargs
+                pg0 = EmbeddedPostgres(**kwargs)
                 # Check if pg0 is already running before we start it
                 was_already_running = await pg0.is_running()
                 self.db_url = await pg0.ensure_running()
@@ -881,11 +885,29 @@ class MemoryEngine(MemoryEngineInterface):
             if not self.db_url:
                 raise ValueError("Database URL is required for migrations")
             logger.info("Running database migrations...")
-            run_migrations(self.db_url)
+            # Use configured database schema for migrations (defaults to "public")
+            run_migrations(self.db_url, schema=get_config().database_schema)
+
+            # Migrate all existing tenant schemas (if multi-tenant)
+            if self._tenant_extension is not None:
+                try:
+                    tenants = await self._tenant_extension.list_tenants()
+                    if tenants:
+                        logger.info(f"Running migrations on {len(tenants)} tenant schemas...")
+                        for tenant in tenants:
+                            schema = tenant.schema
+                            if schema and schema != "public":
+                                try:
+                                    run_migrations(self.db_url, schema=schema)
+                                except Exception as e:
+                                    logger.warning(f"Failed to migrate tenant schema {schema}: {e}")
+                        logger.info("Tenant schema migrations completed")
+                except Exception as e:
+                    logger.warning(f"Failed to run tenant schema migrations: {e}")
 
             # Ensure embedding column dimension matches the model's dimension
             # This is done after migrations and after embeddings.initialize()
-            ensure_embedding_dimension(self.db_url, self.embeddings.dimension)
+            ensure_embedding_dimension(self.db_url, self.embeddings.dimension, schema=get_config().database_schema)
 
         logger.info(f"Connecting to PostgreSQL at {self.db_url}")
 
@@ -1169,15 +1191,15 @@ class MemoryEngine(MemoryEngineInterface):
             context: Context about when/why this memory was formed
             event_date: When the event occurred (defaults to now)
             document_id: Optional document ID for tracking (always upserts if document already exists)
-            fact_type_override: Override fact type ('world', 'experience', 'opinion')
-            confidence_score: Confidence score for opinions (0.0 to 1.0)
+            fact_type_override: Override fact type ('world', 'experience')
+            confidence_score: Confidence score (0.0 to 1.0)
             request_context: Request context for authentication.
 
         Returns:
             List of created unit IDs
         """
         # Build content dict
-        content_dict: RetainContentDict = {"content": content, "context": context}  # type: ignore[typeddict-item] - building incrementally
+        content_dict: RetainContentDict = {"content": content, "context": context}
         if event_date:
             content_dict["event_date"] = event_date
         if document_id:
@@ -1225,8 +1247,8 @@ class MemoryEngine(MemoryEngineInterface):
                 - "document_id" (optional): Document ID for this specific content item
             document_id: **DEPRECATED** - Use "document_id" key in each content dict instead.
                         Applies the same document_id to ALL content items that don't specify their own.
-            fact_type_override: Override fact type for all facts ('world', 'experience', 'opinion')
-            confidence_score: Confidence score for opinions (0.0 to 1.0)
+            fact_type_override: Override fact type for all facts ('world', 'experience')
+            confidence_score: Confidence score (0.0 to 1.0)
             return_usage: If True, returns tuple of (unit_ids, TokenUsage). Default False for backward compatibility.
 
         Returns:
@@ -1548,16 +1570,16 @@ class MemoryEngine(MemoryEngineInterface):
         if fact_type is None:
             fact_type = list(VALID_RECALL_FACT_TYPES)
 
-        # Validate fact types early
+        # Filter out 'opinion' early (deprecated, silently ignore)
+        fact_type = [ft for ft in fact_type if ft != "opinion"]
+
+        # Validate fact types
         invalid_types = set(fact_type) - VALID_RECALL_FACT_TYPES
         if invalid_types:
             raise ValueError(
                 f"Invalid fact type(s): {', '.join(sorted(invalid_types))}. "
                 f"Must be one of: {', '.join(sorted(VALID_RECALL_FACT_TYPES))}"
             )
-
-        # Filter out 'opinion' - opinions are no longer returned from recall
-        fact_type = [ft for ft in fact_type if ft != "opinion"]
         if not fact_type:
             # All requested types were opinions - return empty result
             return RecallResultModel(results=[], entities={}, chunks={})
@@ -2213,44 +2235,15 @@ class MemoryEngine(MemoryEngineInterface):
                     )
                 top_results_dicts.append(result_dict)
 
-            # Get entities for each fact if include_entities is requested
-            fact_entity_map = {}  # unit_id -> list of (entity_id, entity_name)
-            if include_entities and top_scored:
-                unit_ids = [uuid.UUID(sr.id) for sr in top_scored]
-                if unit_ids:
-                    async with acquire_with_retry(pool) as entity_conn:
-                        entity_rows = await entity_conn.fetch(
-                            f"""
-                            SELECT ue.unit_id, e.id as entity_id, e.canonical_name
-                            FROM {fq_table("unit_entities")} ue
-                            JOIN {fq_table("entities")} e ON ue.entity_id = e.id
-                            WHERE ue.unit_id = ANY($1::uuid[])
-                            """,
-                            unit_ids,
-                        )
-                        for row in entity_rows:
-                            unit_id = str(row["unit_id"])
-                            if unit_id not in fact_entity_map:
-                                fact_entity_map[unit_id] = []
-                            fact_entity_map[unit_id].append(
-                                {"entity_id": str(row["entity_id"]), "canonical_name": row["canonical_name"]}
-                            )
-
             # Convert results to MemoryFact objects
             memory_facts = []
             for result_dict in top_results_dicts:
-                result_id = str(result_dict.get("id"))
-                # Get entity names for this fact
-                entity_names = None
-                if include_entities and result_id in fact_entity_map:
-                    entity_names = [e["canonical_name"] for e in fact_entity_map[result_id]]
-
                 memory_facts.append(
                     MemoryFact(
-                        id=result_id,
+                        id=str(result_dict.get("id")),
                         text=result_dict.get("text"),
                         fact_type=result_dict.get("fact_type", "world"),
-                        entities=entity_names,
+                        entities=None,  # Entity observations removed
                         context=result_dict.get("context"),
                         occurred_start=result_dict.get("occurred_start"),
                         occurred_end=result_dict.get("occurred_end"),
@@ -2261,38 +2254,12 @@ class MemoryEngine(MemoryEngineInterface):
                     )
                 )
 
-            # Fetch entity observations if requested
+            # Entity observations removed - always set to None
             entities_dict = None
-            total_entity_tokens = 0
-            total_chunk_tokens = 0
-            if include_entities and fact_entity_map:
-                # Collect unique entities in order of fact relevance (preserving order from top_scored)
-                # Use a list to maintain order, but track seen entities to avoid duplicates
-                entities_ordered = []  # list of (entity_id, entity_name) tuples
-                seen_entity_ids = set()
-
-                # Iterate through facts in relevance order
-                for sr in top_scored:
-                    unit_id = sr.id
-                    if unit_id in fact_entity_map:
-                        for entity in fact_entity_map[unit_id]:
-                            entity_id = entity["entity_id"]
-                            entity_name = entity["canonical_name"]
-                            if entity_id not in seen_entity_ids:
-                                entities_ordered.append((entity_id, entity_name))
-                                seen_entity_ids.add(entity_id)
-
-                # Return entities with empty observations (summaries now live in mental models)
-                entities_dict = {}
-                for entity_id, entity_name in entities_ordered:
-                    entities_dict[entity_name] = EntityState(
-                        entity_id=entity_id,
-                        canonical_name=entity_name,
-                        observations=[],  # Mental models provide this now
-                    )
 
             # Fetch chunks if requested
             chunks_dict = None
+            total_chunk_tokens = 0
             if include_chunks and top_scored:
                 from .response_models import ChunkInfo
 
@@ -2361,7 +2328,6 @@ class MemoryEngine(MemoryEngineInterface):
             # Log final recall stats
             total_time = time.time() - recall_start
             num_chunks = len(chunks_dict) if chunks_dict else 0
-            num_entities = len(entities_dict) if entities_dict else 0
             # Include wait times in log if significant
             wait_parts = []
             if semaphore_wait > 0.01:
@@ -2370,7 +2336,7 @@ class MemoryEngine(MemoryEngineInterface):
                 wait_parts.append(f"conn={max_conn_wait:.3f}s")
             wait_info = f" | waits: {', '.join(wait_parts)}" if wait_parts else ""
             log_buffer.append(
-                f"[RECALL {recall_id}] Complete: {len(top_scored)} facts ({total_tokens} tok), {num_chunks} chunks ({total_chunk_tokens} tok), {num_entities} entities ({total_entity_tokens} tok) | {fact_type_summary} | {total_time:.3f}s{wait_info}"
+                f"[RECALL {recall_id}] Complete: {len(top_scored)} facts ({total_tokens} tok), {num_chunks} chunks ({total_chunk_tokens} tok) | {fact_type_summary} | {total_time:.3f}s{wait_info}"
             )
             if not quiet:
                 logger.info("\n" + "\n".join(log_buffer))
@@ -2764,7 +2730,7 @@ class MemoryEngine(MemoryEngineInterface):
             param_count += 1
             units = await conn.fetch(
                 f"""
-                SELECT id, text, event_date, context, occurred_start, occurred_end, mentioned_at, document_id, chunk_id, fact_type, tags, created_at, proof_count
+                SELECT id, text, event_date, context, occurred_start, occurred_end, mentioned_at, document_id, chunk_id, fact_type, tags, created_at, proof_count, source_memory_ids
                 FROM {fq_table("memory_units")}
                 {where_clause}
                 ORDER BY mentioned_at DESC NULLS LAST, event_date DESC
@@ -2777,7 +2743,18 @@ class MemoryEngine(MemoryEngineInterface):
             # Get links, filtering to only include links between units of the selected agent
             # Use DISTINCT ON with LEAST/GREATEST to deduplicate bidirectional links
             unit_ids = [row["id"] for row in units]
-            if unit_ids:
+            unit_id_set = set(unit_ids)
+
+            # Collect source memory IDs from observations
+            source_memory_ids = []
+            for unit in units:
+                if unit["source_memory_ids"]:
+                    source_memory_ids.extend(unit["source_memory_ids"])
+            source_memory_ids = list(set(source_memory_ids))  # Deduplicate
+
+            # Fetch links involving both visible units AND source memories
+            all_relevant_ids = unit_ids + source_memory_ids
+            if all_relevant_ids:
                 links = await conn.fetch(
                     f"""
                     SELECT DISTINCT ON (LEAST(ml.from_unit_id, ml.to_unit_id), GREATEST(ml.from_unit_id, ml.to_unit_id), ml.link_type, COALESCE(ml.entity_id, '00000000-0000-0000-0000-000000000000'::uuid))
@@ -2788,14 +2765,69 @@ class MemoryEngine(MemoryEngineInterface):
                         e.canonical_name as entity_name
                     FROM {fq_table("memory_links")} ml
                     LEFT JOIN {fq_table("entities")} e ON ml.entity_id = e.id
-                    WHERE ml.from_unit_id = ANY($1::uuid[]) AND ml.to_unit_id = ANY($1::uuid[])
+                    WHERE ml.from_unit_id = ANY($1::uuid[]) OR ml.to_unit_id = ANY($1::uuid[])
                     ORDER BY LEAST(ml.from_unit_id, ml.to_unit_id), GREATEST(ml.from_unit_id, ml.to_unit_id), ml.link_type, COALESCE(ml.entity_id, '00000000-0000-0000-0000-000000000000'::uuid), ml.weight DESC
                 """,
-                    unit_ids,
+                    all_relevant_ids,
                 )
             else:
                 links = []
 
+            # Copy links from source memories to observations
+            # Observations inherit links from their source memories via source_memory_ids
+            # Build a map from source_id to observation_ids
+            source_to_observations = {}
+            for unit in units:
+                if unit["source_memory_ids"]:
+                    for source_id in unit["source_memory_ids"]:
+                        if source_id not in source_to_observations:
+                            source_to_observations[source_id] = []
+                        source_to_observations[source_id].append(unit["id"])
+
+            copied_links = []
+            for link in links:
+                from_id = link["from_unit_id"]
+                to_id = link["to_unit_id"]
+
+                # Get observations that should inherit this link
+                from_observations = source_to_observations.get(from_id, [])
+                to_observations = source_to_observations.get(to_id, [])
+
+                # If from_id is a source memory, copy links to its observations
+                if from_observations:
+                    for obs_id in from_observations:
+                        # Only include if the target is visible
+                        if to_id in unit_id_set or to_observations:
+                            target = to_observations[0] if to_observations and to_id not in unit_id_set else to_id
+                            if target in unit_id_set:
+                                copied_links.append(
+                                    {
+                                        "from_unit_id": obs_id,
+                                        "to_unit_id": target,
+                                        "link_type": link["link_type"],
+                                        "weight": link["weight"],
+                                        "entity_name": link["entity_name"],
+                                    }
+                                )
+
+                # If to_id is a source memory, copy links to its observations
+                if to_observations and from_id in unit_id_set:
+                    for obs_id in to_observations:
+                        copied_links.append(
+                            {
+                                "from_unit_id": from_id,
+                                "to_unit_id": obs_id,
+                                "link_type": link["link_type"],
+                                "weight": link["weight"],
+                                "entity_name": link["entity_name"],
+                            }
+                        )
+
+            # Keep only direct links between visible nodes
+            direct_links = [
+                link for link in links if link["from_unit_id"] in unit_id_set and link["to_unit_id"] in unit_id_set
+            ]
+
             # Get entity information
             unit_entities = await conn.fetch(f"""
                 SELECT ue.unit_id, e.canonical_name
@@ -2813,6 +2845,18 @@ class MemoryEngine(MemoryEngineInterface):
                 entity_map[unit_id] = []
             entity_map[unit_id].append(entity_name)
 
+        # For observations, inherit entities from source memories
+        for unit in units:
+            if unit["source_memory_ids"] and unit["id"] not in entity_map:
+                # Collect entities from all source memories
+                source_entities = []
+                for source_id in unit["source_memory_ids"]:
+                    if source_id in entity_map:
+                        source_entities.extend(entity_map[source_id])
+                if source_entities:
+                    # Deduplicate while preserving order
+                    entity_map[unit["id"]] = list(dict.fromkeys(source_entities))
+
         # Build nodes
         nodes = []
         for row in units:
@@ -2846,14 +2890,15 @@ class MemoryEngine(MemoryEngineInterface):
                 }
             )
 
-        # Build edges
+        # Build edges (combine direct links and copied links from sources)
         edges = []
-        for row in links:
+        all_links = direct_links + copied_links
+        for row in all_links:
             from_id = str(row["from_unit_id"])
             to_id = str(row["to_unit_id"])
             link_type = row["link_type"]
             weight = row["weight"]
-            entity_name = row["entity_name"]
+            entity_name = row.get("entity_name")
 
             # Color by link type
             if link_type == "temporal":
@@ -3465,7 +3510,6 @@ class MemoryEngine(MemoryEngineInterface):
             ReflectResult containing:
                 - text: Plain text answer
                 - based_on: Empty dict (agent retrieves facts dynamically)
-                - new_opinions: Empty list
                 - structured_output: None (not yet supported for agentic reflect)
         """
         # Use cached LLM config
@@ -3790,7 +3834,6 @@ class MemoryEngine(MemoryEngineInterface):
         result = ReflectResult(
             text=agent_result.text,
             based_on=based_on,
-            new_opinions=[],  # Learnings stored as mental models
             structured_output=agent_result.structured_output,
             usage=usage,
             tool_trace=tool_trace_result,
@@ -3819,32 +3862,6 @@ class MemoryEngine(MemoryEngineInterface):
 
         return result
 
-    async def get_entity_observations(
-        self,
-        bank_id: str,
-        entity_id: str,
-        *,
-        limit: int = 10,
-        request_context: "RequestContext",
-    ) -> list[Any]:
-        """
-        Get observations for an entity.
-
-        NOTE: Entity observations/summaries have been moved to mental models.
-        This method returns an empty list. Use mental models for entity summaries.
-
-        Args:
-            bank_id: bank IDentifier
-            entity_id: Entity UUID to get observations for
-            limit: Ignored (kept for backwards compatibility)
-            request_context: Request context for authentication.
-
-        Returns:
-            Empty list (observations now in mental models)
-        """
-        await self._authenticate_tenant(request_context)
-        return []
-
     async def list_entities(
         self,
         bank_id: str,
@@ -4031,36 +4048,6 @@ class MemoryEngine(MemoryEngineInterface):
         await self._authenticate_tenant(request_context)
         return EntityState(entity_id=entity_id, canonical_name=entity_name, observations=[])
 
-    async def regenerate_entity_observations(
-        self,
-        bank_id: str,
-        entity_id: str,
-        entity_name: str,
-        *,
-        version: str | None = None,
-        conn=None,
-        request_context: "RequestContext",
-    ) -> list[str]:
-        """
-        Regenerate observations for an entity.
-
-        NOTE: Entity observations/summaries have been moved to mental models.
-        This method is now a no-op and returns an empty list.
-
-        Args:
-            bank_id: bank IDentifier
-            entity_id: Entity UUID
-            entity_name: Canonical name of the entity
-            version: Entity's last_seen timestamp when task was created (for deduplication)
-            conn: Optional database connection (ignored)
-            request_context: Request context for authentication.
-
-        Returns:
-            Empty list (observations now in mental models)
-        """
-        await self._authenticate_tenant(request_context)
-        return []
-
     # =========================================================================
     # Statistics & Operations (for HTTP API layer)
     # =========================================================================
@@ -4171,9 +4158,6 @@ class MemoryEngine(MemoryEngineInterface):
         if not entity_row:
             return None
 
-        # Get observations for the entity
-        observations = await self.get_entity_observations(bank_id, entity_id, limit=20, request_context=request_context)
-
         return {
             "id": str(entity_row["id"]),
             "canonical_name": entity_row["canonical_name"],
@@ -4181,7 +4165,7 @@ class MemoryEngine(MemoryEngineInterface):
             "first_seen": entity_row["first_seen"].isoformat() if entity_row["first_seen"] else None,
             "last_seen": entity_row["last_seen"].isoformat() if entity_row["last_seen"] else None,
             "metadata": entity_row["metadata"] or {},
-            "observations": observations,
+            "observations": [],
         }
 
     def _parse_observations(self, observations_raw: list):

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

@@ -58,6 +58,7 @@ def _normalize_tool_name(name: str) -> str:
     - 'functions.done' (OpenAI-style prefix)
     - 'call=functions.done' (some models)
     - 'call=done' (some models)
+    - 'done<|channel|>commentary' (malformed special tokens appended)
 
     Returns the normalized tool name (e.g., 'done', 'recall', etc.)
     """
@@ -69,6 +70,11 @@ def _normalize_tool_name(name: str) -> str:
     if name.startswith("functions."):
         name = name[len("functions.") :]
 
+    # Handle malformed special tokens appended to tool name
+    # e.g., 'done<|channel|>commentary' -> 'done'
+    if "<|" in name:
+        name = name.split("<|")[0]
+
     return name
 
 

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

@@ -69,7 +69,7 @@ async def tool_search_mental_models(
         next_param += 1
 
     if exclude_ids:
-        filters += f" AND id != ALL(${next_param}::uuid[])"
+        filters += f" AND id != ALL(${next_param}::text[])"
         params.append(exclude_ids)
         next_param += 1
 

hindsight-api/hindsight_api/engine/response_models.py

@@ -263,7 +263,6 @@ class ReflectResult(BaseModel):
                         }
                     ],
                 },
-                "new_opinions": ["Machine learning has great potential in healthcare"],
                 "structured_output": {"summary": "ML in healthcare", "confidence": 0.9},
                 "usage": {"input_tokens": 1500, "output_tokens": 500, "total_tokens": 2000},
             }
@@ -272,9 +271,8 @@ class ReflectResult(BaseModel):
 
     text: str = Field(description="The formulated answer text")
     based_on: dict[str, Any] = Field(
-        description="Facts used to formulate the answer, organized by type (world, experience, opinion, mental_models, directives)"
+        description="Facts used to formulate the answer, organized by type (world, experience, mental_models, directives)"
     )
-    new_opinions: list[str] = Field(default_factory=list, description="List of newly formed opinions during reflection")
     structured_output: dict[str, Any] | None = Field(
         default=None,
         description="Structured output parsed according to the provided response schema. Only present when response_schema was provided.",
@@ -297,24 +295,6 @@ class ReflectResult(BaseModel):
     )
 
 
-class Opinion(BaseModel):
-    """
-    An opinion with confidence score.
-
-    Opinions represent the bank's formed perspectives on topics,
-    with a confidence level indicating strength of belief.
-    """
-
-    model_config = ConfigDict(
-        json_schema_extra={
-            "example": {"text": "Machine learning has great potential in healthcare", "confidence": 0.85}
-        }
-    )
-
-    text: str = Field(description="The opinion text")
-    confidence: float = Field(description="Confidence score between 0.0 and 1.0")
-
-
 class EntityObservation(BaseModel):
     """
     An observation about an entity.

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

@@ -693,7 +693,6 @@ async def _extract_facts_from_chunk(
     context: str,
     llm_config: "LLMConfig",
     agent_name: str = None,
-    extract_opinions: bool = False,
 ) -> tuple[list[dict[str, str]], TokenUsage]:
     """
     Extract facts from a single chunk (internal helper for parallel processing).
@@ -707,17 +706,9 @@ async def _extract_facts_from_chunk(
 
     logger = logging.getLogger(__name__)
 
-    memory_bank_context = f"\n- Your name: {agent_name}" if agent_name and extract_opinions else ""
-
-    # Determine which fact types to extract based on the flag
+    # Determine which fact types to extract
     # Note: We use "assistant" in the prompt but convert to "bank" for storage
-    if extract_opinions:
-        # Opinion extraction uses a separate prompt (not this one)
-        fact_types_instruction = "Extract ONLY 'opinion' type facts (formed opinions, beliefs, and perspectives). DO NOT extract 'world' or 'assistant' facts."
-    else:
-        fact_types_instruction = (
-            "Extract ONLY 'world' and 'assistant' type facts. DO NOT extract opinions - those are extracted separately."
-        )
+    fact_types_instruction = "Extract ONLY 'world' and 'assistant' type facts."
 
     # Check config for extraction mode and causal link extraction
     config = get_config()
@@ -770,7 +761,6 @@ async def _extract_facts_from_chunk(
     # Format event_date with day of week for better temporal reasoning
     event_date_formatted = event_date.strftime("%A, %B %d, %Y")  # e.g., "Monday, June 10, 2024"
     user_message = f"""Extract facts from the following text chunk.
-{memory_bank_context}
 
 Chunk: {chunk_index + 1}/{total_chunks}
 Event Date: {event_date_formatted} ({event_date.isoformat()})
@@ -782,12 +772,28 @@ Text:
     usage = TokenUsage()  # Track cumulative usage across retries
     for attempt in range(max_retries):
         try:
+            # Use retain-specific overrides if set, otherwise fall back to global LLM config
+            max_retries = (
+                config.retain_llm_max_retries if config.retain_llm_max_retries is not None else config.llm_max_retries
+            )
+            initial_backoff = (
+                config.retain_llm_initial_backoff
+                if config.retain_llm_initial_backoff is not None
+                else config.llm_initial_backoff
+            )
+            max_backoff = (
+                config.retain_llm_max_backoff if config.retain_llm_max_backoff is not None else config.llm_max_backoff
+            )
+
             extraction_response_json, call_usage = await llm_config.call(
                 messages=[{"role": "system", "content": prompt}, {"role": "user", "content": user_message}],
                 response_format=response_schema,
                 scope="memory_extract_facts",
                 temperature=0.1,
                 max_completion_tokens=config.retain_max_completion_tokens,
+                max_retries=max_retries,
+                initial_backoff=initial_backoff,
+                max_backoff=max_backoff,
                 skip_validation=True,  # Get raw JSON, we'll validate leniently
                 return_usage=True,
             )
@@ -1013,7 +1019,6 @@ async def _extract_facts_with_auto_split(
     context: str,
     llm_config: LLMConfig,
     agent_name: str = None,
-    extract_opinions: bool = False,
 ) -> tuple[list[dict[str, str]], TokenUsage]:
     """
     Extract facts from a chunk with automatic splitting if output exceeds token limits.
@@ -1029,7 +1034,6 @@ async def _extract_facts_with_auto_split(
         context: Context about the conversation/document
         llm_config: LLM configuration to use
         agent_name: Optional agent name (memory owner)
-        extract_opinions: If True, extract ONLY opinions. If False, extract world and agent facts (no opinions)
 
     Returns:
         Tuple of (facts list, token usage) extracted from the chunk (possibly from sub-chunks)
@@ -1048,7 +1052,6 @@ async def _extract_facts_with_auto_split(
             context=context,
             llm_config=llm_config,
             agent_name=agent_name,
-            extract_opinions=extract_opinions,
         )
     except OutputTooLongError:
         # Output exceeded token limits - split the chunk in half and retry
@@ -1093,7 +1096,6 @@ async def _extract_facts_with_auto_split(
                 context=context,
                 llm_config=llm_config,
                 agent_name=agent_name,
-                extract_opinions=extract_opinions,
             ),
             _extract_facts_with_auto_split(
                 chunk=second_half,
@@ -1103,7 +1105,6 @@ async def _extract_facts_with_auto_split(
                 context=context,
                 llm_config=llm_config,
                 agent_name=agent_name,
-                extract_opinions=extract_opinions,
             ),
         ]
 
@@ -1127,7 +1128,6 @@ async def extract_facts_from_text(
     llm_config: LLMConfig,
     agent_name: str,
     context: str = "",
-    extract_opinions: bool = False,
 ) -> tuple[list[Fact], list[tuple[str, int]], TokenUsage]:
     """
     Extract semantic facts from conversational or narrative text using LLM.
@@ -1144,7 +1144,6 @@ async def extract_facts_from_text(
         context: Context about the conversation/document
         llm_config: LLM configuration to use
         agent_name: Agent name (memory owner)
-        extract_opinions: If True, extract ONLY opinions. If False, extract world and bank facts (no opinions)
 
     Returns:
         Tuple of (facts, chunks, usage) where:
@@ -1172,7 +1171,6 @@ async def extract_facts_from_text(
             context=context,
             llm_config=llm_config,
             agent_name=agent_name,
-            extract_opinions=extract_opinions,
         )
         for i, chunk in enumerate(chunks)
     ]
@@ -1204,7 +1202,7 @@ SECONDS_PER_FACT = 10
 
 
 async def extract_facts_from_contents(
-    contents: list[RetainContent], llm_config, agent_name: str, extract_opinions: bool = False
+    contents: list[RetainContent], llm_config, agent_name: str
 ) -> tuple[list[ExtractedFactType], list[ChunkMetadata], TokenUsage]:
     """
     Extract facts from multiple content items in parallel.
@@ -1219,7 +1217,6 @@ async def extract_facts_from_contents(
         contents: List of RetainContent objects to process
         llm_config: LLM configuration for fact extraction
         agent_name: Name of the agent (for agent-related fact detection)
-        extract_opinions: If True, extract only opinions; otherwise world/bank facts
 
     Returns:
         Tuple of (extracted_facts, chunks_metadata, usage)
@@ -1238,7 +1235,6 @@ async def extract_facts_from_contents(
             context=item.context,
             llm_config=llm_config,
             agent_name=agent_name,
-            extract_opinions=extract_opinions,
         )
         fact_extraction_tasks.append(task)
 

hindsight-api/hindsight_api/engine/retain/orchestrator.py

@@ -101,11 +101,8 @@ async def retain_batch(
 
     # Step 1: Extract facts from all contents
     step_start = time.time()
-    extract_opinions = fact_type_override == "opinion"
 
-    extracted_facts, chunks, usage = await fact_extraction.extract_facts_from_contents(
-        contents, llm_config, agent_name, extract_opinions
-    )
+    extracted_facts, chunks, usage = await fact_extraction.extract_facts_from_contents(contents, llm_config, agent_name)
     log_buffer.append(
         f"[1] Extract facts: {len(extracted_facts)} facts, {len(chunks)} chunks from {len(contents)} contents in {time.time() - step_start:.3f}s"
     )

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

@@ -155,7 +155,6 @@ class LinkExpansionRetriever(GraphRetriever):
                 all_seeds.extend(temporal_seeds)
 
             if not all_seeds:
-                logger.info("[LinkExpansion] No seeds found, returning empty results")
                 return [], timings
 
             seed_ids = list({s.id for s in all_seeds})

hindsight-api/hindsight_api/engine/utils.py

@@ -19,7 +19,6 @@ async def extract_facts(
     context: str = "",
     llm_config: "LLMConfig" = None,
     agent_name: str = None,
-    extract_opinions: bool = False,
 ) -> tuple[list["Fact"], list[tuple[str, int]]]:
     """
     Extract semantic facts from text using LLM.
@@ -36,7 +35,6 @@ async def extract_facts(
         context: Context about the conversation/document
         llm_config: LLM configuration to use
         agent_name: Optional agent name to help identify agent-related facts
-        extract_opinions: If True, extract ONLY opinions. If False, extract world and agent facts (no opinions)
 
     Returns:
         Tuple of (facts, chunks) where:
@@ -55,7 +53,6 @@ async def extract_facts(
         context=context,
         llm_config=llm_config,
         agent_name=agent_name,
-        extract_opinions=extract_opinions,
     )
 
     if not facts:

hindsight-api/hindsight_api/extensions/builtin/tenant.py

@@ -1,5 +1,6 @@
 """Built-in tenant extension implementations."""
 
+from hindsight_api.config import get_config
 from hindsight_api.extensions.tenant import AuthenticationError, Tenant, TenantContext, TenantExtension
 from hindsight_api.models import RequestContext
 
@@ -10,11 +11,13 @@ class ApiKeyTenantExtension(TenantExtension):
 
     This is a simple implementation that:
     1. Validates the API key matches HINDSIGHT_API_TENANT_API_KEY
-    2. Returns 'public' as the schema for all authenticated requests
+    2. Returns the configured schema (HINDSIGHT_API_DATABASE_SCHEMA, default 'public')
+       for all authenticated requests
 
     Configuration:
         HINDSIGHT_API_TENANT_EXTENSION=hindsight_api.extensions.builtin.tenant:ApiKeyTenantExtension
         HINDSIGHT_API_TENANT_API_KEY=your-secret-key
+        HINDSIGHT_API_DATABASE_SCHEMA=your-schema (optional, defaults to 'public')
 
     For multi-tenant setups with separate schemas per tenant, implement a custom
     TenantExtension that looks up the schema based on the API key or token claims.
@@ -27,11 +30,11 @@ class ApiKeyTenantExtension(TenantExtension):
             raise ValueError("HINDSIGHT_API_TENANT_API_KEY is required when using ApiKeyTenantExtension")
 
     async def authenticate(self, context: RequestContext) -> TenantContext:
-        """Validate API key and return public schema context."""
+        """Validate API key and return configured schema context."""
         if context.api_key != self.expected_api_key:
             raise AuthenticationError("Invalid API key")
-        return TenantContext(schema_name="public")
+        return TenantContext(schema_name=get_config().database_schema)
 
     async def list_tenants(self) -> list[Tenant]:
-        """Return public schema for single-tenant setup."""
-        return [Tenant(schema="public")]
+        """Return configured schema for single-tenant setup."""
+        return [Tenant(schema=get_config().database_schema)]

hindsight-api/hindsight_api/main.py

@@ -170,31 +170,56 @@ def main():
     if args.log_level != config.log_level:
         config = HindsightConfig(
             database_url=config.database_url,
+            database_schema=config.database_schema,
             llm_provider=config.llm_provider,
             llm_api_key=config.llm_api_key,
             llm_model=config.llm_model,
             llm_base_url=config.llm_base_url,
             llm_max_concurrent=config.llm_max_concurrent,
+            llm_max_retries=config.llm_max_retries,
+            llm_initial_backoff=config.llm_initial_backoff,
+            llm_max_backoff=config.llm_max_backoff,
             llm_timeout=config.llm_timeout,
+            llm_vertexai_project_id=config.llm_vertexai_project_id,
+            llm_vertexai_region=config.llm_vertexai_region,
+            llm_vertexai_service_account_key=config.llm_vertexai_service_account_key,
             retain_llm_provider=config.retain_llm_provider,
             retain_llm_api_key=config.retain_llm_api_key,
             retain_llm_model=config.retain_llm_model,
             retain_llm_base_url=config.retain_llm_base_url,
+            retain_llm_max_concurrent=config.retain_llm_max_concurrent,
+            retain_llm_max_retries=config.retain_llm_max_retries,
+            retain_llm_initial_backoff=config.retain_llm_initial_backoff,
+            retain_llm_max_backoff=config.retain_llm_max_backoff,
+            retain_llm_timeout=config.retain_llm_timeout,
             reflect_llm_provider=config.reflect_llm_provider,
             reflect_llm_api_key=config.reflect_llm_api_key,
             reflect_llm_model=config.reflect_llm_model,
             reflect_llm_base_url=config.reflect_llm_base_url,
+            reflect_llm_max_concurrent=config.reflect_llm_max_concurrent,
+            reflect_llm_max_retries=config.reflect_llm_max_retries,
+            reflect_llm_initial_backoff=config.reflect_llm_initial_backoff,
+            reflect_llm_max_backoff=config.reflect_llm_max_backoff,
+            reflect_llm_timeout=config.reflect_llm_timeout,
             consolidation_llm_provider=config.consolidation_llm_provider,
             consolidation_llm_api_key=config.consolidation_llm_api_key,
             consolidation_llm_model=config.consolidation_llm_model,
             consolidation_llm_base_url=config.consolidation_llm_base_url,
+            consolidation_llm_max_concurrent=config.consolidation_llm_max_concurrent,
+            consolidation_llm_max_retries=config.consolidation_llm_max_retries,
+            consolidation_llm_initial_backoff=config.consolidation_llm_initial_backoff,
+            consolidation_llm_max_backoff=config.consolidation_llm_max_backoff,
+            consolidation_llm_timeout=config.consolidation_llm_timeout,
             embeddings_provider=config.embeddings_provider,
             embeddings_local_model=config.embeddings_local_model,
+            embeddings_local_force_cpu=config.embeddings_local_force_cpu,
             embeddings_tei_url=config.embeddings_tei_url,
             embeddings_openai_base_url=config.embeddings_openai_base_url,
             embeddings_cohere_base_url=config.embeddings_cohere_base_url,
             reranker_provider=config.reranker_provider,
             reranker_local_model=config.reranker_local_model,
+            reranker_local_force_cpu=config.reranker_local_force_cpu,
+            reranker_local_max_concurrent=config.reranker_local_max_concurrent,
             reranker_tei_url=config.reranker_tei_url,
             reranker_tei_batch_size=config.reranker_tei_batch_size,
             reranker_tei_max_concurrent=config.reranker_tei_max_concurrent,
@@ -214,9 +239,9 @@ def main():
             retain_extract_causal_links=config.retain_extract_causal_links,
             retain_extraction_mode=config.retain_extraction_mode,
             retain_custom_instructions=config.retain_custom_instructions,
-            retain_observations_async=config.retain_observations_async,
             enable_observations=config.enable_observations,
             consolidation_batch_size=config.consolidation_batch_size,
+            consolidation_max_tokens=config.consolidation_max_tokens,
             skip_llm_verification=config.skip_llm_verification,
             lazy_reranker=config.lazy_reranker,
             run_migrations_on_startup=config.run_migrations_on_startup,
@@ -228,8 +253,9 @@ def main():
             worker_id=config.worker_id,
             worker_poll_interval_ms=config.worker_poll_interval_ms,
             worker_max_retries=config.worker_max_retries,
-            worker_batch_size=config.worker_batch_size,
             worker_http_port=config.worker_http_port,
+            worker_max_slots=config.worker_max_slots,
+            worker_consolidation_max_slots=config.worker_consolidation_max_slots,
             reflect_max_iterations=config.reflect_max_iterations,
             mental_model_refresh_concurrency=config.mental_model_refresh_concurrency,
         )
@@ -341,6 +367,7 @@ def main():
     # Start idle checker in daemon mode
     if idle_middleware is not None:
         # Start the idle checker in a background thread with its own event loop
+        import logging
         import threading
 
         def run_idle_checker():
@@ -351,12 +378,12 @@ def main():
                 loop = asyncio.new_event_loop()
                 asyncio.set_event_loop(loop)
                 loop.run_until_complete(idle_middleware._check_idle())
-            except Exception:
-                pass
+            except Exception as e:
+                logging.error(f"Idle checker error: {e}", exc_info=True)
 
         threading.Thread(target=run_idle_checker, daemon=True).start()
 
-    uvicorn.run(**uvicorn_config)  # type: ignore[invalid-argument-type] - dict kwargs
+    uvicorn.run(**uvicorn_config)
 
 
 if __name__ == "__main__":

hindsight-api/hindsight_api/mcp_tools.py

@@ -32,6 +32,9 @@ class MCPToolsConfig:
     # How to resolve bank_id for operations
     bank_id_resolver: Callable[[], str | None]
 
+    # How to resolve API key for tenant auth (optional)
+    api_key_resolver: Callable[[], str | None] | None = None
+
     # Whether to include bank_id as a parameter on tools (for multi-bank support)
     include_bank_id_param: bool = False
 
@@ -46,6 +49,16 @@ class MCPToolsConfig:
     retain_fire_and_forget: bool = False  # If True, use asyncio.create_task pattern
 
 
+def _get_request_context(config: MCPToolsConfig) -> RequestContext:
+    """Create RequestContext with API key from resolver if available.
+
+    This enables tenant auth to work with MCP tools by propagating
+    the Bearer token from the MCP middleware to the memory engine.
+    """
+    api_key = config.api_key_resolver() if config.api_key_resolver else None
+    return RequestContext(api_key=api_key)
+
+
 def parse_timestamp(timestamp: str) -> datetime | None:
     """Parse an ISO format timestamp string.
 
@@ -155,12 +168,14 @@ def _register_retain(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig)
                 if error:
                     return {"status": "error", "message": error}
 
+                request_context = _get_request_context(config)
+
                 async def _retain():
                     try:
                         await memory.retain_batch_async(
                             bank_id=target_bank,
                             contents=[content_dict],
-                            request_context=RequestContext(),
+                            request_context=request_context,
                         )
                     except Exception as e:
                         logger.error(f"Error storing memory: {e}", exc_info=True)
@@ -196,16 +211,17 @@ def _register_retain(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig)
                         return f"Error: {error}"
 
                     contents = [content_dict]
+                    request_context = _get_request_context(config)
                     if async_processing:
                         result = await memory.submit_async_retain(
-                            bank_id=target_bank, contents=contents, request_context=RequestContext()
+                            bank_id=target_bank, contents=contents, request_context=request_context
                         )
                         return f"Memory queued for background processing (operation_id: {result.get('operation_id', 'N/A')})"
                     else:
                         await memory.retain_batch_async(
                             bank_id=target_bank,
                             contents=contents,
-                            request_context=RequestContext(),
+                            request_context=request_context,
                         )
                         return f"Memory stored successfully in bank '{target_bank}'"
                 except Exception as e:
@@ -237,12 +253,14 @@ def _register_retain(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig)
             if error:
                 return {"status": "error", "message": error}
 
+            request_context = _get_request_context(config)
+
             async def _retain():
                 try:
                     await memory.retain_batch_async(
                         bank_id=target_bank,
                         contents=[content_dict],
-                        request_context=RequestContext(),
+                        request_context=request_context,
                     )
                 except Exception as e:
                     logger.error(f"Error storing memory: {e}", exc_info=True)
@@ -280,7 +298,7 @@ def _register_recall(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig)
                     fact_type=list(VALID_RECALL_FACT_TYPES),
                     budget=Budget.HIGH,
                     max_tokens=max_tokens,
-                    request_context=RequestContext(),
+                    request_context=_get_request_context(config),
                 )
 
                 return recall_result.model_dump_json(indent=2)
@@ -311,7 +329,7 @@ def _register_recall(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig)
                     fact_type=list(VALID_RECALL_FACT_TYPES),
                     budget=Budget.HIGH,
                     max_tokens=max_tokens,
-                    request_context=RequestContext(),
+                    request_context=_get_request_context(config),
                 )
 
                 return recall_result.model_dump()
@@ -370,7 +388,7 @@ def _register_reflect(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig
                     query=query,
                     budget=budget_enum,
                     context=context,
-                    request_context=RequestContext(),
+                    request_context=_get_request_context(config),
                 )
 
                 return reflect_result.model_dump_json(indent=2)
@@ -423,7 +441,7 @@ def _register_reflect(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsConfig
                     query=query,
                     budget=budget_enum,
                     context=context,
-                    request_context=RequestContext(),
+                    request_context=_get_request_context(config),
                 )
 
                 return reflect_result.model_dump()
@@ -447,7 +465,7 @@ def _register_list_banks(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsCon
             JSON list of banks with their IDs, names, dispositions, and missions.
         """
         try:
-            banks = await memory.list_banks(request_context=RequestContext())
+            banks = await memory.list_banks(request_context=_get_request_context(config))
             return json.dumps({"banks": banks}, indent=2)
         except Exception as e:
             logger.error(f"Error listing banks: {e}", exc_info=True)
@@ -471,8 +489,9 @@ def _register_create_bank(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsCo
             mission: Optional mission describing who the agent is and what they're trying to accomplish
         """
         try:
+            request_context = _get_request_context(config)
             # get_bank_profile auto-creates bank if it doesn't exist
-            profile = await memory.get_bank_profile(bank_id, request_context=RequestContext())
+            profile = await memory.get_bank_profile(bank_id, request_context=request_context)
 
             # Update name/mission if provided
             if name is not None or mission is not None:
@@ -480,10 +499,10 @@ def _register_create_bank(mcp: FastMCP, memory: MemoryEngine, config: MCPToolsCo
                     bank_id,
                     name=name,
                     mission=mission,
-                    request_context=RequestContext(),
+                    request_context=request_context,
                 )
                 # Fetch updated profile
-                profile = await memory.get_bank_profile(bank_id, request_context=RequestContext())
+                profile = await memory.get_bank_profile(bank_id, request_context=request_context)
 
             # Serialize disposition if it's a Pydantic model
             if "disposition" in profile and hasattr(profile["disposition"], "model_dump"):

hindsight-api/hindsight_api/metrics.py

@@ -189,7 +189,7 @@ class MetricsCollectorBase:
         Args:
             provider: LLM provider name (openai, anthropic, gemini, groq, ollama, lmstudio)
             model: Model name
-            scope: Scope identifier (e.g., "memory", "reflect", "entity_observation")
+            scope: Scope identifier (e.g., "memory", "reflect", "consolidation")
             duration: Call duration in seconds
             input_tokens: Number of input/prompt tokens
             output_tokens: Number of output/completion tokens
@@ -321,7 +321,7 @@ class MetricsCollector(MetricsCollectorBase):
                 pass
 
         Args:
-            operation: Operation name (retain, recall, reflect, entity_observation)
+            operation: Operation name (retain, recall, reflect, consolidation)
             bank_id: Memory bank ID
             source: Source of the operation (api, reflect, internal)
             budget: Optional budget level (low, mid, high)
@@ -371,7 +371,7 @@ class MetricsCollector(MetricsCollectorBase):
         Args:
             provider: LLM provider name (openai, anthropic, gemini, groq, ollama, lmstudio)
             model: Model name
-            scope: Scope identifier (e.g., "memory", "reflect", "entity_observation")
+            scope: Scope identifier (e.g., "memory", "reflect", "consolidation")
             duration: Call duration in seconds
             input_tokens: Number of input/prompt tokens
             output_tokens: Number of output/completion tokens

hindsight-api/hindsight_api/pg0.py

@@ -40,7 +40,7 @@ class EmbeddedPostgres:
             # Only set port if explicitly specified
             if self.port is not None:
                 kwargs["port"] = self.port
-            self._pg0 = Pg0(**kwargs)  # type: ignore[invalid-argument-type] - dict kwargs
+            self._pg0 = Pg0(**kwargs)
         return self._pg0
 
     async def start(self, max_retries: int = 5, retry_delay: float = 4.0) -> str:

hindsight-api/hindsight_api/worker/main.py

@@ -124,12 +124,6 @@ def main():
         default=config.worker_poll_interval_ms,
         help=f"Poll interval in milliseconds (default: {config.worker_poll_interval_ms}, env: HINDSIGHT_API_WORKER_POLL_INTERVAL_MS)",
     )
-    parser.add_argument(
-        "--batch-size",
-        type=int,
-        default=config.worker_batch_size,
-        help=f"Tasks to claim per poll (default: {config.worker_batch_size}, env: HINDSIGHT_API_WORKER_BATCH_SIZE)",
-    )
     parser.add_argument(
         "--max-retries",
         type=int,
@@ -168,8 +162,9 @@ def main():
 
     print(f"Starting Hindsight Worker: {args.worker_id}")
     print(f"  Poll interval: {args.poll_interval}ms")
-    print(f"  Batch size: {args.batch_size}")
     print(f"  Max retries: {args.max_retries}")
+    print(f"  Max slots: {config.worker_max_slots}")
+    print(f"  Consolidation max slots: {config.worker_consolidation_max_slots}")
     print(f"  HTTP server: {args.http_host}:{args.http_port}")
     print()
 
@@ -183,21 +178,25 @@ def main():
 
         from ..extensions import TenantExtension, load_extension
 
+        # Load tenant extension BEFORE creating MemoryEngine so it can
+        # set correct schema context during task execution. Without this,
+        # _authenticate_tenant sees no extension and resets schema to "public",
+        # causing worker writes to land in the wrong schema.
+        tenant_extension = load_extension("TENANT", TenantExtension)
+
         # Initialize MemoryEngine
         # Workers use SyncTaskBackend because they execute tasks directly,
         # they don't need to store tasks (they poll from DB)
         memory = MemoryEngine(
             run_migrations=False,  # Workers don't run migrations
             task_backend=SyncTaskBackend(),
+            tenant_extension=tenant_extension,
         )
 
         await memory.initialize()
 
         print(f"Database connected: {config.database_url}")
 
-        # Load tenant extension for dynamic schema discovery
-        tenant_extension = load_extension("TENANT", TenantExtension)
-
         if tenant_extension:
             print("Tenant extension loaded - schemas will be discovered dynamically on each poll")
         else:
@@ -209,9 +208,10 @@ def main():
             worker_id=args.worker_id,
             executor=memory.execute_task,
             poll_interval_ms=args.poll_interval,
-            batch_size=args.batch_size,
             max_retries=args.max_retries,
             tenant_extension=tenant_extension,
+            max_slots=config.worker_max_slots,
+            consolidation_max_slots=config.worker_consolidation_max_slots,
         )
 
         # Create the HTTP app for metrics/health

hindsight-api/hindsight_api/worker/poller.py

@@ -57,10 +57,11 @@ class WorkerPoller:
         worker_id: str,
         executor: Callable[[dict[str, Any]], Awaitable[None]],
         poll_interval_ms: int = 500,
-        batch_size: int = 10,
         max_retries: int = 3,
         schema: str | None = None,
         tenant_extension: "TenantExtension | None" = None,
+        max_slots: int = 10,
+        consolidation_max_slots: int = 2,
     ):
         """
         Initialize the worker poller.
@@ -70,28 +71,32 @@ class WorkerPoller:
             worker_id: Unique identifier for this worker
             executor: Async function to execute tasks (typically MemoryEngine.execute_task)
             poll_interval_ms: Interval between polls when no tasks found (milliseconds)
-            batch_size: Maximum number of tasks to claim per poll cycle
             max_retries: Maximum retry attempts before marking task as failed
             schema: Database schema for single-tenant support (ignored if tenant_extension is set)
             tenant_extension: Extension for dynamic multi-tenant discovery. If set, list_tenants()
                             is called on each poll cycle to discover schemas dynamically.
+            max_slots: Maximum concurrent tasks per worker
+            consolidation_max_slots: Maximum concurrent consolidation tasks per worker
         """
         self._pool = pool
         self._worker_id = worker_id
         self._executor = executor
         self._poll_interval_ms = poll_interval_ms
-        self._batch_size = batch_size
         self._max_retries = max_retries
         self._schema = schema
         self._tenant_extension = tenant_extension
+        self._max_slots = max_slots
+        self._consolidation_max_slots = consolidation_max_slots
         self._shutdown = asyncio.Event()
         self._current_tasks: set[asyncio.Task] = set()
         self._in_flight_count = 0
         self._in_flight_lock = asyncio.Lock()
         self._last_progress_log = 0.0
         self._tasks_completed_since_log = 0
-        # Track active tasks locally: operation_id -> (op_type, bank_id, schema)
-        self._active_tasks: dict[str, tuple[str, str, str | None]] = {}
+        # Track active tasks locally: operation_id -> (op_type, bank_id, schema, asyncio.Task)
+        self._active_tasks: dict[str, tuple[str, str, str | None, asyncio.Task]] = {}
+        # Track in-flight tasks by operation type
+        self._in_flight_by_type: dict[str, int] = {}
 
     async def _get_schemas(self) -> list[str | None]:
         """Get list of schemas to poll. Returns [None] for public schema."""
@@ -102,59 +107,114 @@ class WorkerPoller:
         # Single schema mode
         return [self._schema]
 
+    async def _get_available_slots(self) -> tuple[int, int]:
+        """
+        Calculate available slots for claiming tasks.
+
+        Returns:
+            (total_available, consolidation_available) tuple
+        """
+        async with self._in_flight_lock:
+            total_in_flight = self._in_flight_count
+            consolidation_in_flight = self._in_flight_by_type.get("consolidation", 0)
+
+        total_available = max(0, self._max_slots - total_in_flight)
+        consolidation_available = max(0, self._consolidation_max_slots - consolidation_in_flight)
+
+        return total_available, consolidation_available
+
+    async def wait_for_active_tasks(self, timeout: float = 10.0) -> bool:
+        """
+        Wait for all active background tasks to complete (test helper).
+
+        This is a test-only utility that allows tests to synchronize with
+        fire-and-forget background tasks without using sleep().
+
+        Args:
+            timeout: Maximum time to wait in seconds
+
+        Returns:
+            True if all tasks completed, False if timeout was reached
+        """
+        start_time = asyncio.get_event_loop().time()
+        while True:
+            async with self._in_flight_lock:
+                if self._in_flight_count == 0:
+                    return True
+
+            elapsed = asyncio.get_event_loop().time() - start_time
+            if elapsed >= timeout:
+                return False
+
+            # Short sleep to avoid busy-waiting
+            await asyncio.sleep(0.01)
+
     async def claim_batch(self) -> list[ClaimedTask]:
         """
-        Claim up to batch_size pending tasks atomically across all tenant schemas.
+        Claim pending tasks atomically across all tenant schemas,
+        respecting slot limits (total and consolidation).
 
         Uses FOR UPDATE SKIP LOCKED to ensure no conflicts with other workers.
 
-        For consolidation tasks specifically, skips pending tasks if there's already
-        a processing consolidation for the same bank (to avoid duplicate work).
-
-        If tenant_extension is configured, dynamically discovers schemas on each call.
-
         Returns:
             List of ClaimedTask objects containing operation_id, task_dict, and schema
         """
+        # Calculate available slots
+        total_available, consolidation_available = await self._get_available_slots()
+
+        if total_available <= 0:
+            return []
+
         schemas = await self._get_schemas()
         all_tasks: list[ClaimedTask] = []
-        remaining_batch = self._batch_size
+        remaining_total = total_available
+        remaining_consolidation = consolidation_available
 
         for schema in schemas:
-            if remaining_batch <= 0:
+            if remaining_total <= 0:
                 break
 
-            tasks = await self._claim_batch_for_schema(schema, remaining_batch)
+            tasks = await self._claim_batch_for_schema(schema, remaining_total, remaining_consolidation)
+
+            # Update remaining slots based on what was claimed
+            for task in tasks:
+                op_type = task.task_dict.get("operation_type", "unknown")
+                if op_type == "consolidation":
+                    remaining_consolidation -= 1
+
             all_tasks.extend(tasks)
-            remaining_batch -= len(tasks)
+            remaining_total -= len(tasks)
 
         return all_tasks
 
-    async def _claim_batch_for_schema(self, schema: str | None, limit: int) -> list[ClaimedTask]:
-        """Claim tasks from a specific schema."""
+    async def _claim_batch_for_schema(
+        self, schema: str | None, limit: int, consolidation_limit: int
+    ) -> list[ClaimedTask]:
+        """Claim tasks from a specific schema respecting slot limits."""
+        try:
+            return await self._claim_batch_for_schema_inner(schema, limit, consolidation_limit)
+        except Exception as e:
+            logger.warning(f"Worker {self._worker_id} failed to claim tasks for schema {schema or 'public'}: {e}")
+            return []
+
+    async def _claim_batch_for_schema_inner(
+        self, schema: str | None, limit: int, consolidation_limit: int
+    ) -> list[ClaimedTask]:
+        """Inner implementation for claiming tasks from a specific schema with slot limits."""
         table = fq_table("async_operations", schema)
 
         async with self._pool.acquire() as conn:
             async with conn.transaction():
-                # Select and lock pending tasks
-                # For consolidation: skip if same bank already has one processing
-                rows = await conn.fetch(
+                # Strategy: Claim non-consolidation tasks first, then consolidation up to limit
+
+                # 1. Claim non-consolidation tasks (up to limit)
+                non_consolidation_rows = await conn.fetch(
                     f"""
                     SELECT operation_id, task_payload
-                    FROM {table} AS pending
-                    WHERE status = 'pending' AND task_payload IS NOT NULL
-                    AND (
-                        -- Non-consolidation tasks: always claimable
-                        operation_type != 'consolidation'
-                        OR
-                        -- Consolidation: only if no other consolidation processing for same bank
-                        NOT EXISTS (
-                            SELECT 1 FROM {table} AS processing
-                            WHERE processing.bank_id = pending.bank_id
-                            AND processing.operation_type = 'consolidation'
-                            AND processing.status = 'processing'
-                        )
-                    )
+                    FROM {table}
+                    WHERE status = 'pending'
+                      AND task_payload IS NOT NULL
+                      AND operation_type != 'consolidation'
                     ORDER BY created_at
                     LIMIT $1
                     FOR UPDATE SKIP LOCKED
@@ -162,11 +222,39 @@ class WorkerPoller:
                     limit,
                 )
 
-                if not rows:
+                claimed_count = len(non_consolidation_rows)
+                remaining_limit = limit - claimed_count
+
+                # 2. Claim consolidation tasks (up to consolidation_limit and remaining_limit)
+                consolidation_rows = []
+                if consolidation_limit > 0 and remaining_limit > 0:
+                    consolidation_rows = await conn.fetch(
+                        f"""
+                        SELECT operation_id, task_payload
+                        FROM {table} AS pending
+                        WHERE status = 'pending'
+                          AND task_payload IS NOT NULL
+                          AND operation_type = 'consolidation'
+                          AND NOT EXISTS (
+                              SELECT 1 FROM {table} AS processing
+                              WHERE processing.bank_id = pending.bank_id
+                                AND processing.operation_type = 'consolidation'
+                                AND processing.status = 'processing'
+                          )
+                        ORDER BY created_at
+                        LIMIT $1
+                        FOR UPDATE SKIP LOCKED
+                        """,
+                        min(consolidation_limit, remaining_limit),
+                    )
+
+                all_rows = non_consolidation_rows + consolidation_rows
+
+                if not all_rows:
                     return []
 
                 # Claim the tasks by updating status and worker_id
-                operation_ids = [row["operation_id"] for row in rows]
+                operation_ids = [row["operation_id"] for row in all_rows]
                 await conn.execute(
                     f"""
                     UPDATE {table}
@@ -184,7 +272,7 @@ class WorkerPoller:
                         task_dict=json.loads(row["task_payload"]),
                         schema=schema,
                     )
-                    for row in rows
+                    for row in all_rows
                 ]
 
     async def _mark_completed(self, operation_id: str, schema: str | None):
@@ -250,18 +338,43 @@ class WorkerPoller:
             logger.warning(f"Task {operation_id} failed, will retry (attempt {retry_count + 1}/{self._max_retries})")
 
     async def execute_task(self, task: ClaimedTask):
-        """Execute a single task and update its status."""
+        """Execute a single task as a background job (fire-and-forget)."""
         task_type = task.task_dict.get("type", "unknown")
+        operation_type = task.task_dict.get("operation_type", "unknown")
         bank_id = task.task_dict.get("bank_id", "unknown")
 
+        # Create background task
+        bg_task = asyncio.create_task(self._execute_task_inner(task))
+
         # Track this task as active
         async with self._in_flight_lock:
-            self._active_tasks[task.operation_id] = (task_type, bank_id, task.schema)
+            self._active_tasks[task.operation_id] = (task_type, bank_id, task.schema, bg_task)
+            self._in_flight_count += 1
+            self._in_flight_by_type[operation_type] = self._in_flight_by_type.get(operation_type, 0) + 1
+
+        # Add cleanup callback
+        bg_task.add_done_callback(lambda _: asyncio.create_task(self._cleanup_task(task.operation_id, operation_type)))
+
+    async def _cleanup_task(self, operation_id: str, operation_type: str):
+        """Remove task from tracking after completion."""
+        async with self._in_flight_lock:
+            if operation_id in self._active_tasks:
+                self._active_tasks.pop(operation_id, None)
+                self._in_flight_count -= 1
+                count = self._in_flight_by_type.get(operation_type, 0)
+                if count > 0:
+                    self._in_flight_by_type[operation_type] = count - 1
+                    if self._in_flight_by_type[operation_type] == 0:
+                        del self._in_flight_by_type[operation_type]
+
+    async def _execute_task_inner(self, task: ClaimedTask):
+        """Inner task execution with error handling."""
+        task_type = task.task_dict.get("type", "unknown")
+        bank_id = task.task_dict.get("bank_id", "unknown")
 
         try:
             schema_info = f", schema={task.schema}" if task.schema else ""
             logger.debug(f"Executing task {task.operation_id} (type={task_type}, bank={bank_id}{schema_info})")
-            # Pass schema to executor so it can set the correct context
             if task.schema:
                 task.task_dict["_schema"] = task.schema
             await self._executor(task.task_dict)
@@ -271,10 +384,6 @@ class WorkerPoller:
             error_msg = f"{type(e).__name__}: {e}\n{traceback.format_exc()}"
             logger.error(f"Task {task.operation_id} failed: {e}")
             await self._retry_or_fail(task.operation_id, error_msg, task.schema)
-        finally:
-            # Remove from active tasks
-            async with self._in_flight_lock:
-                self._active_tasks.pop(task.operation_id, None)
 
     async def recover_own_tasks(self) -> int:
         """
@@ -293,20 +402,23 @@ class WorkerPoller:
         total_count = 0
 
         for schema in schemas:
-            table = fq_table("async_operations", schema)
+            try:
+                table = fq_table("async_operations", schema)
 
-            result = await self._pool.execute(
-                f"""
-                UPDATE {table}
-                SET status = 'pending', worker_id = NULL, claimed_at = NULL, updated_at = now()
-                WHERE status = 'processing' AND worker_id = $1
-                """,
-                self._worker_id,
-            )
+                result = await self._pool.execute(
+                    f"""
+                    UPDATE {table}
+                    SET status = 'pending', worker_id = NULL, claimed_at = NULL, updated_at = now()
+                    WHERE status = 'processing' AND worker_id = $1
+                    """,
+                    self._worker_id,
+                )
 
-            # Parse "UPDATE N" to get count
-            count = int(result.split()[-1]) if result else 0
-            total_count += count
+                # Parse "UPDATE N" to get count
+                count = int(result.split()[-1]) if result else 0
+                total_count += count
+            except Exception as e:
+                logger.warning(f"Worker {self._worker_id} failed to recover tasks for schema {schema or 'public'}: {e}")
 
         if total_count > 0:
             logger.info(f"Worker {self._worker_id} recovered {total_count} stale tasks from previous run")
@@ -314,59 +426,59 @@ class WorkerPoller:
 
     async def run(self):
         """
-        Main polling loop.
+        Main polling loop with fire-and-forget task execution.
 
-        Continuously polls for pending tasks, claims them, and executes them
-        until shutdown is signaled.
-
-        If tenant_extension is configured, dynamically discovers schemas on each poll.
+        Continuously polls for pending tasks, spawns them as background tasks,
+        and immediately continues polling (up to slot limits).
         """
-        # Recover any tasks from a previous crash before starting
         await self.recover_own_tasks()
 
-        logger.info(f"Worker {self._worker_id} starting polling loop")
+        logger.info(
+            f"Worker {self._worker_id} starting polling loop "
+            f"(max_slots={self._max_slots}, consolidation_max_slots={self._consolidation_max_slots})"
+        )
 
         while not self._shutdown.is_set():
             try:
-                # Claim a batch of tasks (across all tenant schemas if configured)
+                # Claim a batch of tasks (respecting slot limits)
                 tasks = await self.claim_batch()
 
                 if tasks:
                     # Log batch info
                     task_types: dict[str, int] = {}
                     schemas_seen: set[str | None] = set()
+                    consolidation_count = 0
                     for task in tasks:
                         t = task.task_dict.get("type", "unknown")
+                        op_type = task.task_dict.get("operation_type", "unknown")
                         task_types[t] = task_types.get(t, 0) + 1
                         schemas_seen.add(task.schema)
+                        if op_type == "consolidation":
+                            consolidation_count += 1
+
                     types_str = ", ".join(f"{k}:{v}" for k, v in task_types.items())
                     schemas_str = ", ".join(s or "public" for s in schemas_seen)
                     logger.info(
-                        f"Worker {self._worker_id} claimed {len(tasks)} tasks: {types_str} (schemas: {schemas_str})"
+                        f"Worker {self._worker_id} claimed {len(tasks)} tasks "
+                        f"({consolidation_count} consolidation): {types_str} (schemas: {schemas_str})"
                     )
 
-                    # Track in-flight tasks
-                    async with self._in_flight_lock:
-                        self._in_flight_count += len(tasks)
+                    # Spawn tasks as background jobs (fire-and-forget)
+                    for task in tasks:
+                        await self.execute_task(task)
 
-                    # Execute tasks concurrently
-                    try:
-                        await asyncio.gather(
-                            *[self.execute_task(task) for task in tasks],
-                            return_exceptions=True,
-                        )
-                    finally:
-                        async with self._in_flight_lock:
-                            self._in_flight_count -= len(tasks)
-                else:
-                    # No tasks found, wait before polling again
-                    try:
-                        await asyncio.wait_for(
-                            self._shutdown.wait(),
-                            timeout=self._poll_interval_ms / 1000,
-                        )
-                    except asyncio.TimeoutError:
-                        pass  # Normal timeout, continue polling
+                    # Continue immediately to claim more tasks (if slots available)
+                    continue
+
+                # No tasks claimed (either no pending tasks or slots full)
+                # Wait before polling again
+                try:
+                    await asyncio.wait_for(
+                        self._shutdown.wait(),
+                        timeout=self._poll_interval_ms / 1000,
+                    )
+                except asyncio.TimeoutError:
+                    pass  # Normal timeout, continue polling
 
                 # Log progress stats periodically
                 await self._log_progress_if_due()
@@ -397,15 +509,27 @@ class WorkerPoller:
         while asyncio.get_event_loop().time() - start_time < timeout:
             async with self._in_flight_lock:
                 in_flight = self._in_flight_count
+                active_task_objects = [task_info[3] for task_info in self._active_tasks.values()]
 
             if in_flight == 0:
                 logger.info(f"Worker {self._worker_id} graceful shutdown complete")
                 return
 
             logger.info(f"Worker {self._worker_id} waiting for {in_flight} in-flight tasks")
-            await asyncio.sleep(0.5)
 
-        logger.warning(f"Worker {self._worker_id} shutdown timeout after {timeout}s")
+            # Wait for at least one task to complete
+            if active_task_objects:
+                done, _ = await asyncio.wait(active_task_objects, timeout=0.5, return_when=asyncio.FIRST_COMPLETED)
+            else:
+                await asyncio.sleep(0.5)
+
+        logger.warning(f"Worker {self._worker_id} shutdown timeout after {timeout}s, cancelling remaining tasks")
+
+        # Cancel remaining tasks
+        async with self._in_flight_lock:
+            for operation_id, (_, _, _, bg_task) in list(self._active_tasks.items()):
+                if not bg_task.done():
+                    bg_task.cancel()
 
     async def _log_progress_if_due(self):
         """Log progress stats every PROGRESS_LOG_INTERVAL seconds."""
@@ -416,14 +540,19 @@ class WorkerPoller:
         self._last_progress_log = now
 
         try:
-            # Get local active tasks (this worker only)
+            # Get local active tasks
             async with self._in_flight_lock:
                 in_flight = self._in_flight_count
-                active_tasks = dict(self._active_tasks)  # Copy to avoid holding lock
+                in_flight_by_type = dict(self._in_flight_by_type)
+                active_tasks = dict(self._active_tasks)
 
-            # Build local processing breakdown grouped by (op_type, bank_id)
+            consolidation_count = in_flight_by_type.get("consolidation", 0)
+            available_slots = self._max_slots - in_flight
+            available_consolidation_slots = self._consolidation_max_slots - consolidation_count
+
+            # Build local processing breakdown
             task_groups: dict[tuple[str, str], int] = {}
-            for op_type, bank_id, _ in active_tasks.values():
+            for op_type, bank_id, _, _ in active_tasks.values():
                 key = (op_type, bank_id)
                 task_groups[key] = task_groups.get(key, 0) + 1
 
@@ -432,7 +561,7 @@ class WorkerPoller:
             if len(processing_info) > 10:
                 processing_str += f" +{len(processing_info) - 10} more"
 
-            # Get global stats from DB across all schemas
+            # Get global stats from DB
             schemas = await self._get_schemas()
             global_pending = 0
             all_worker_counts: dict[str, int] = {}
@@ -444,7 +573,6 @@ class WorkerPoller:
                     row = await conn.fetchrow(f"SELECT COUNT(*) as count FROM {table} WHERE status = 'pending'")
                     global_pending += row["count"] if row else 0
 
-                    # Get processing breakdown by worker
                     worker_rows = await conn.fetch(
                         f"""
                         SELECT worker_id, COUNT(*) as count
@@ -457,7 +585,6 @@ class WorkerPoller:
                         wid = wr["worker_id"] or "unknown"
                         all_worker_counts[wid] = all_worker_counts.get(wid, 0) + wr["count"]
 
-            # Format other workers' processing counts
             other_workers = []
             for wid, cnt in all_worker_counts.items():
                 if wid != self._worker_id:
@@ -466,7 +593,9 @@ class WorkerPoller:
 
             schemas_str = ", ".join(s or "public" for s in schemas)
             logger.info(
-                f"[WORKER_STATS] worker={self._worker_id} in_flight={in_flight} | "
+                f"[WORKER_STATS] worker={self._worker_id} "
+                f"slots={in_flight}/{self._max_slots} (consolidation={consolidation_count}/{self._consolidation_max_slots}) | "
+                f"available={available_slots} (consolidation={available_consolidation_slots}) | "
                 f"global: pending={global_pending} (schemas: {schemas_str}) | "
                 f"others: {others_str} | "
                 f"my_active: {processing_str}"

hindsight-api/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "hindsight-api"
-version = "0.3.0"
+version = "0.4.2"
 description = "Hindsight: Agent Memory That Works Like Human Memory"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -34,6 +34,7 @@ dependencies = [
     "opentelemetry-exporter-prometheus>=0.41b0",
     "dateparser>=1.2.2",
     "google-genai>=1.0.0",
+    "google-auth>=2.0.0",
     "anthropic>=0.40.0",
     "typer>=0.9.0",
     "cohere>=5.0.0",
@@ -141,6 +142,11 @@ known-third-party = ["alembic"]
 quote-style = "double"
 indent-style = "space"
 
+[tool.uv]
+# Allow uv to search all configured indexes for packages, not just the first one
+# This prevents dependency resolution failures when using pytorch index + PyPI
+index-strategy = "unsafe-best-match"
+
 [tool.ty]
 # Type checking configuration
 # ty is an extremely fast Python type checker from Astral (same team as ruff/uv)

hindsight-api/tests/test_consolidation.py

@@ -346,11 +346,11 @@ class TestConsolidationIntegration:
         or when one directly updates another (e.g., location change).
 
         Given:
-        - "Nicolò lives in Italy"
-        - "Nicolò moved to the US recently" (updates the living location)
+        - "Alex lives in Italy"
+        - "Alex moved to the US recently" (updates the living location)
 
         The second fact should UPDATE the first, not create a separate observation.
-        But unrelated facts like "Nicolò works at Vectorize" should stay separate.
+        But unrelated facts like "Alex works at Vectorize" should stay separate.
         """
         bank_id = f"test-consolidation-merge-{uuid.uuid4().hex[:8]}"
 
@@ -360,14 +360,14 @@ class TestConsolidationIntegration:
         # Retain a memory about living location
         await memory.retain_async(
             bank_id=bank_id,
-            content="Nicolò lives in Italy.",
+            content="Alex lives in Italy.",
             request_context=request_context,
         )
 
         # Retain an unrelated memory (different topic - should NOT merge)
         await memory.retain_async(
             bank_id=bank_id,
-            content="Nicolò works at Vectorize as an engineer.",
+            content="Alex works at Vectorize as an engineer.",
             request_context=request_context,
         )
 
@@ -384,7 +384,7 @@ class TestConsolidationIntegration:
         # Add a memory that UPDATES the living location (should merge with first)
         await memory.retain_async(
             bank_id=bank_id,
-            content="Nicolò recently moved to the United States.",
+            content="Alex recently moved to the United States.",
             request_context=request_context,
         )
 
@@ -485,9 +485,9 @@ class TestConsolidationIntegration:
         they should be merged into ONE observation that captures the change.
 
         Example:
-        - "Nicolò loves pizza"
-        - "Nicolò hates pizza"
-        → Should become: "Nicolò used to love pizza but now hates it" (or similar)
+        - "Alex loves pizza"
+        - "Alex hates pizza"
+        → Should become: "Alex used to love pizza but now hates it" (or similar)
         """
         bank_id = f"test-consolidation-contradict-{uuid.uuid4().hex[:8]}"
 
@@ -497,7 +497,7 @@ class TestConsolidationIntegration:
         # Add initial fact
         await memory.retain_async(
             bank_id=bank_id,
-            content="Nicolò loves pizza.",
+            content="Alex loves pizza.",
             request_context=request_context,
         )
 
@@ -515,7 +515,7 @@ class TestConsolidationIntegration:
         # Add contradicting fact (same person, same topic, opposite sentiment)
         await memory.retain_async(
             bank_id=bank_id,
-            content="Nicolò hates pizza.",
+            content="Alex hates pizza.",
             request_context=request_context,
         )
 
@@ -1897,3 +1897,93 @@ class TestMentalModelRefreshAfterConsolidation:
 
         # Cleanup
         await memory.delete_bank(bank_id, request_context=request_context)
+
+    @pytest.mark.asyncio
+    async def test_graph_endpoint_observations_inherit_links_and_entities(
+        self, memory: MemoryEngine, request_context
+    ):
+        """Test that graph endpoint shows links and entities for observations filtered by type.
+
+        When filtering graph by type=observation:
+        - Observations should inherit links from their source memories
+        - Observations should show entities inherited from source memories
+        - Even when source memories are not visible, their links should be copied to observations
+        """
+        bank_id = f"test-graph-obs-{uuid.uuid4().hex[:8]}"
+
+        # Create the bank
+        await memory.get_bank_profile(bank_id=bank_id, request_context=request_context)
+
+        # Retain content that will create world facts with shared entities
+        # This should create facts that are linked by shared entities
+        await memory.retain_async(
+            bank_id=bank_id,
+            content="Alice works at Google as a software engineer.",
+            request_context=request_context,
+        )
+
+        await memory.retain_async(
+            bank_id=bank_id,
+            content="Bob also works at Google in the sales department.",
+            request_context=request_context,
+        )
+
+        # Wait for consolidation to create observations
+        import asyncio
+
+        await asyncio.sleep(2)
+
+        # Get graph data filtered by observation type only
+        graph_data = await memory.get_graph_data(
+            bank_id=bank_id,
+            fact_type="observation",
+            limit=1000,
+            request_context=request_context,
+        )
+
+        # Should have observations
+        assert graph_data["total_units"] > 0, "Should have observations"
+        assert len(graph_data["nodes"]) > 0, "Should have observation nodes"
+
+        # Verify all nodes are observations
+        for row in graph_data["table_rows"]:
+            assert row["fact_type"] == "observation", f"All nodes should be observations, got {row['fact_type']}"
+
+        # Should have edges (inherited from source memories)
+        # Even though we're only showing observations, they should inherit links from their sources
+        assert len(graph_data["edges"]) > 0, (
+            "Observations should have edges inherited from source memories. "
+            f"Found {len(graph_data['edges'])} edges"
+        )
+
+        # Should have entities (inherited from source memories)
+        observations_with_entities = [
+            row for row in graph_data["table_rows"] if row["entities"] and row["entities"] != "None"
+        ]
+        assert len(observations_with_entities) > 0, (
+            "Observations should inherit entities from source memories. "
+            f"Found {len(observations_with_entities)} observations with entities"
+        )
+
+        # Verify entities contain expected values
+        all_entities = " ".join([row["entities"] for row in graph_data["table_rows"]])
+        assert "Alice" in all_entities or "Bob" in all_entities or "Google" in all_entities, (
+            f"Expected to find Alice, Bob, or Google in entities, got: {all_entities}"
+        )
+
+        # Verify edge types are valid
+        valid_link_types = {"semantic", "temporal", "entity"}
+        for edge in graph_data["edges"]:
+            link_type = edge["data"]["linkType"]
+            assert link_type in valid_link_types, f"Invalid link type: {link_type}"
+
+        # Verify all edges connect visible observation nodes
+        visible_node_ids = {row["id"] for row in graph_data["table_rows"]}
+        for edge in graph_data["edges"]:
+            source_id = edge["data"]["source"]
+            target_id = edge["data"]["target"]
+            assert source_id in visible_node_ids, f"Edge source {source_id[:8]} not in visible nodes"
+            assert target_id in visible_node_ids, f"Edge target {target_id[:8]} not in visible nodes"
+
+        # Cleanup
+        await memory.delete_bank(bank_id, request_context=request_context)

hindsight-api/tests/test_fact_extraction_analysis.py

@@ -58,7 +58,6 @@ async def test_fact_extraction_basic_analysis(llm_config):
         llm_config=llm_config,
         agent_name="test-agent",
         context="Friday Standup meeting",
-        extract_opinions=False,
     )
 
     duration = time.time() - start_time

hindsight-api/tests/test_http_api_integration.py

@@ -1063,3 +1063,38 @@ async def test_retain_async_no_usage(api_client):
 
     # Usage should be None for async operations
     assert result.get("usage") is None, "Async retain should not include usage"
+
+
+@pytest.mark.asyncio
+async def test_version_endpoint_returns_correct_version(api_client):
+    """Test that the /version endpoint returns the correct API version.
+
+    The version should match the __version__ defined in hindsight_api.__init__.py
+    and should not be a hardcoded string.
+    """
+    from hindsight_api import __version__
+
+    # Call the /version endpoint
+    response = await api_client.get("/version")
+    assert response.status_code == 200
+    result = response.json()
+
+    # Verify response structure
+    assert "api_version" in result, "Response should include 'api_version' field"
+    assert "features" in result, "Response should include 'features' field"
+
+    # Verify the version matches the package version
+    assert result["api_version"] == __version__, (
+        f"API version should be {__version__}, got {result['api_version']}"
+    )
+
+    # Verify features field structure
+    features = result["features"]
+    assert "observations" in features
+    assert "mcp" in features
+    assert "worker" in features
+    assert isinstance(features["observations"], bool)
+    assert isinstance(features["mcp"], bool)
+    assert isinstance(features["worker"], bool)
+
+    print(f"Version endpoint returned: api_version={result['api_version']}, features={features}")

hindsight-api/tests/test_mcp_routing.py

@@ -97,3 +97,47 @@ def test_path_parsing_logic():
     bank_id, remaining = parse_path("/my-bank/some/path")
     assert bank_id == "my-bank"
     assert remaining == "/some/path"
+
+
+@pytest.mark.asyncio
+async def test_api_key_context_variable():
+    """Test that API key context variable works correctly."""
+    from hindsight_api.api.mcp import get_current_api_key, _current_api_key
+
+    # Initially None
+    assert get_current_api_key() is None
+
+    # Set and verify
+    token = _current_api_key.set("test-api-key-123")
+    try:
+        assert get_current_api_key() == "test-api-key-123"
+    finally:
+        _current_api_key.reset(token)
+
+    # Back to None after reset
+    assert get_current_api_key() is None
+
+
+@pytest.mark.asyncio
+async def test_mcp_tools_propagate_api_key(mock_memory):
+    """Test that MCP tools propagate API key to RequestContext."""
+    from hindsight_api.api.mcp import create_mcp_server, _current_bank_id, _current_api_key
+
+    mcp_server = create_mcp_server(mock_memory)
+    tools = mcp_server._tool_manager._tools
+
+    # Set both bank_id and api_key context
+    bank_token = _current_bank_id.set("test-bank")
+    api_key_token = _current_api_key.set("test-bearer-token")
+    try:
+        retain_tool = tools["retain"]
+        result = await retain_tool.fn(content="test content", context="test_context", async_processing=False)
+        assert "successfully" in result.lower()
+
+        # Verify the memory was called with request_context containing api_key
+        mock_memory.retain_batch_async.assert_called_once()
+        call_kwargs = mock_memory.retain_batch_async.call_args.kwargs
+        assert call_kwargs["request_context"].api_key == "test-bearer-token"
+    finally:
+        _current_bank_id.reset(bank_token)
+        _current_api_key.reset(api_key_token)

hindsight-api/tests/test_metrics.py

@@ -358,7 +358,7 @@ class TestLLMMetrics:
         collector.record_llm_call(
             provider="gemini",
             model="gemini-pro",
-            scope="entity_observation",
+            scope="memory",
             duration=2.0,
             success=True,
         )
@@ -369,11 +369,11 @@ class TestLLMMetrics:
         assert call_args[0][0] == 1
         assert call_args[0][1]["provider"] == "gemini"
         assert call_args[0][1]["model"] == "gemini-pro"
-        assert call_args[0][1]["scope"] == "entity_observation"
+        assert call_args[0][1]["scope"] == "memory"
 
     def test_record_llm_call_different_scopes(self, collector):
         """Test recording LLM calls with different scopes."""
-        scopes = ["memory", "reflect", "entity_observation", "answer"]
+        scopes = ["memory", "reflect", "consolidation", "answer"]
 
         for scope in scopes:
             collector.llm_duration.record.reset_mock()

hindsight-api/tests/test_multilingual.py

@@ -469,7 +469,6 @@ async def test_mixed_language_entities(memory, request_context):
             budget=Budget.MID,
             max_tokens=1000,
             fact_type=["world"],
-            include_entities=True,
             request_context=request_context,
         )
 

hindsight-api/tests/test_observations.py

@@ -91,156 +91,13 @@ async def test_entity_extraction_on_retain(memory, request_context):
             await conn.execute("DELETE FROM entities WHERE bank_id = $1", bank_id)
 
 
-@pytest.mark.asyncio
-async def test_regenerate_entity_observations(memory, request_context):
-    """
-    Test explicit regeneration of summary for an entity.
-    """
-    bank_id = f"test_regen_obs_{datetime.now(timezone.utc).timestamp()}"
-
-    try:
-        # Store facts about an entity
-        await memory.retain_async(
-            bank_id=bank_id,
-            content="Sarah is a product manager who loves user research and data analysis.",
-            context="work info",
-            event_date=datetime(2024, 1, 15, tzinfo=timezone.utc),
-            request_context=request_context,
-        )
-
-        await memory.wait_for_background_tasks()
-
-        # Find the Sarah entity
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            entity_row = await conn.fetchrow(
-                """
-                SELECT id, canonical_name
-                FROM entities
-                WHERE bank_id = $1 AND LOWER(canonical_name) LIKE '%sarah%'
-                LIMIT 1
-                """,
-                bank_id
-            )
-
-        if entity_row:
-            entity_id = str(entity_row['id'])
-            entity_name = entity_row['canonical_name']
-
-            # Manually regenerate summary (via observations API for backwards compat)
-            created_ids = await memory.regenerate_entity_observations(
-                bank_id=bank_id,
-                entity_id=entity_id,
-                entity_name=entity_name,
-                request_context=request_context,
-            )
-
-            print(f"\n=== Regenerated Summary ===")
-            print(f"Created {len(created_ids)} summary for {entity_name}")
-
-            # Get entity state
-            state = await memory.get_entity_state(
-                bank_id, entity_id, entity_name, request_context=request_context
-            )
-            for obs in state.observations:
-                print(f"  - {obs.text}")
-
-            # Verify summary was created
-            if len(created_ids) > 0:
-                assert len(state.observations) == 1, "Should have exactly 1 observation (the summary)"
-                print(f"Summary regenerated successfully")
-            else:
-                print(f"Note: No summary was regenerated")
-
-        else:
-            print(f"Note: No 'Sarah' entity was extracted")
-
-    finally:
-        # Cleanup
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            await conn.execute("DELETE FROM memory_units WHERE bank_id = $1", bank_id)
-            await conn.execute("DELETE FROM entities WHERE bank_id = $1", bank_id)
-
-
-@pytest.mark.asyncio
-async def test_entity_state_retrieval(memory, request_context):
-    """
-    Test retrieving entity state with facts.
-    """
-    bank_id = f"test_entity_state_{datetime.now(timezone.utc).timestamp()}"
-
-    try:
-        # Store facts
-        await memory.retain_async(
-            bank_id=bank_id,
-            content="Alice works at Google as a senior software engineer.",
-            context="work info",
-            event_date=datetime(2024, 1, 15, tzinfo=timezone.utc),
-            request_context=request_context,
-        )
-        await memory.retain_async(
-            bank_id=bank_id,
-            content="Alice loves hiking and outdoor photography.",
-            context="hobbies",
-            event_date=datetime(2024, 1, 16, tzinfo=timezone.utc),
-            request_context=request_context,
-        )
-
-        # Find the Alice entity
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            entity_row = await conn.fetchrow(
-                """
-                SELECT id, canonical_name
-                FROM entities
-                WHERE bank_id = $1 AND LOWER(canonical_name) LIKE '%alice%'
-                LIMIT 1
-                """,
-                bank_id
-            )
-
-        assert entity_row is not None, "Alice entity should have been extracted"
-
-        entity_id = str(entity_row['id'])
-        entity_name = entity_row['canonical_name']
-
-        # Check fact count
-        async with pool.acquire() as conn:
-            fact_count = await conn.fetchval(
-                "SELECT COUNT(*) FROM unit_entities WHERE entity_id = $1",
-                entity_row['id']
-            )
-
-        print(f"\n=== Entity State Test ===")
-        print(f"Entity: {entity_name} (id: {entity_id})")
-        print(f"Linked facts: {fact_count}")
-
-        # Get entity state
-        state = await memory.get_entity_state(
-            bank_id, entity_id, entity_name, request_context=request_context
-        )
-
-        assert state.entity_id == entity_id
-        assert state.canonical_name == entity_name
-        print(f"Entity state retrieved successfully")
-
-    finally:
-        # Cleanup
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            await conn.execute("DELETE FROM memory_units WHERE bank_id = $1", bank_id)
-            await conn.execute("DELETE FROM entities WHERE bank_id = $1", bank_id)
-
-
 @pytest.mark.asyncio
 async def test_search_with_include_entities(memory, request_context):
     """
-    Test that search with include_entities=True returns entity information.
+    Test that recall accepts include_entities parameter for backwards compatibility.
 
-    This test verifies that:
-    1. Entities are extracted after retain
-    2. Entity info is returned in recall results with include_entities=True
+    Note: Entity observations have been deprecated. This test verifies the parameter
+    is still accepted without errors.
     """
     bank_id = f"test_search_ent_{datetime.now(timezone.utc).timestamp()}"
 
@@ -249,10 +106,6 @@ async def test_search_with_include_entities(memory, request_context):
         contents = [
             "Alice is a data scientist who works on recommendation systems at Netflix.",
             "Alice presented her research at the ML conference last month.",
-            "Alice is an expert in deep learning and neural networks.",
-            "Alice graduated from Stanford with a PhD in Computer Science.",
-            "Alice leads a team of 5 data scientists at Netflix.",
-            "Alice published a paper on collaborative filtering algorithms.",
         ]
 
         for i, content in enumerate(contents):
@@ -267,7 +120,7 @@ async def test_search_with_include_entities(memory, request_context):
         # Wait for background tasks
         await memory.wait_for_background_tasks()
 
-        # Search with include_entities=True
+        # Search with include_entities=True (should be accepted for backwards compatibility)
         result = await memory.recall_async(
             bank_id=bank_id,
             query="What does Alice do?",
@@ -279,98 +132,9 @@ async def test_search_with_include_entities(memory, request_context):
             request_context=request_context,
         )
 
-        print(f"\n=== Search Results ===")
-        print(f"Found {len(result.results)} facts")
-        for fact in result.results:
-            print(f"  - {fact.text}")
-            if fact.entities:
-                print(f"    Entities: {', '.join(fact.entities)}")
-
-        # Verify results
+        # Verify recall works
         assert len(result.results) > 0, "Should find some facts"
-
-        # Check if entities are included in facts
-        facts_with_entities = [f for f in result.results if f.entities]
-        assert len(facts_with_entities) > 0, "Some facts should have entity information"
-        print(f"{len(facts_with_entities)} facts have entity information")
-
-        # Check if entity info is returned
-        if result.entities:
-            print(f"Entity info included for {len(result.entities)} entities")
-
-            # Verify Alice entity is in results
-            alice_found = False
-            for name, state in result.entities.items():
-                assert state.canonical_name == name, "Entity canonical_name should match key"
-                assert state.entity_id, "Entity should have an ID"
-                if "alice" in name.lower():
-                    alice_found = True
-                    print(f"Alice entity found: {name}")
-
-            assert alice_found, "Alice entity should be in recall results"
-
-    finally:
-        # Cleanup
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            await conn.execute("DELETE FROM memory_units WHERE bank_id = $1", bank_id)
-            await conn.execute("DELETE FROM entities WHERE bank_id = $1", bank_id)
-
-
-@pytest.mark.asyncio
-async def test_get_entity_state(memory, request_context):
-    """
-    Test getting the full state of an entity.
-    """
-    bank_id = f"test_entity_state_{datetime.now(timezone.utc).timestamp()}"
-
-    try:
-        # Store facts
-        await memory.retain_async(
-            bank_id=bank_id,
-            content="Bob is a frontend developer who specializes in React and TypeScript.",
-            context="work info",
-            event_date=datetime(2024, 1, 15, tzinfo=timezone.utc),
-            request_context=request_context,
-        )
-
-        await memory.wait_for_background_tasks()
-
-        # Find entity
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            entity_row = await conn.fetchrow(
-                """
-                SELECT id, canonical_name
-                FROM entities
-                WHERE bank_id = $1 AND LOWER(canonical_name) LIKE '%bob%'
-                LIMIT 1
-                """,
-                bank_id
-            )
-
-        if entity_row:
-            entity_id = str(entity_row['id'])
-            entity_name = entity_row['canonical_name']
-
-            # Get entity state
-            state = await memory.get_entity_state(
-                bank_id=bank_id,
-                entity_id=entity_id,
-                entity_name=entity_name,
-                limit=10,
-                request_context=request_context,
-            )
-
-            print(f"\n=== Entity State for {entity_name} ===")
-            print(f"Entity ID: {state.entity_id}")
-            print(f"Canonical Name: {state.canonical_name}")
-            print(f"Observations: {len(state.observations)}")
-            for obs in state.observations:
-                print(f"  - {obs.text}")
-
-            assert state.entity_id == entity_id, "Entity ID should match"
-            assert state.canonical_name == entity_name, "Canonical name should match"
+        print(f"Found {len(result.results)} facts")
 
     finally:
         # Cleanup

hindsight-api/tests/test_per_operation_llm_config.py

@@ -275,3 +275,88 @@ class TestReflectUsesReflectLLMConfig:
 
         # Verify it's different from the retain config
         assert engine._reflect_llm_config.model != engine._retain_llm_config.model
+
+
+class TestRetryAndBackoffConfiguration:
+    """Test retry and backoff configuration options."""
+
+    def test_global_retry_backoff_config_defaults(self):
+        """Test that global retry/backoff settings have correct defaults."""
+        from hindsight_api.config import get_config
+
+        config = get_config()
+
+        # Verify global defaults
+        assert config.llm_max_retries == 10
+        assert config.llm_initial_backoff == 1.0
+        assert config.llm_max_backoff == 60.0
+
+    def test_per_operation_retry_backoff_config_from_env(self):
+        """Test that per-operation retry/backoff settings are loaded from environment."""
+        from hindsight_api.config import clear_config_cache
+
+        # Set per-operation overrides
+        os.environ["HINDSIGHT_API_RETAIN_LLM_MAX_RETRIES"] = "3"
+        os.environ["HINDSIGHT_API_RETAIN_LLM_INITIAL_BACKOFF"] = "2.0"
+        os.environ["HINDSIGHT_API_RETAIN_LLM_MAX_BACKOFF"] = "120.0"
+        os.environ["HINDSIGHT_API_REFLECT_LLM_MAX_RETRIES"] = "5"
+        os.environ["HINDSIGHT_API_REFLECT_LLM_INITIAL_BACKOFF"] = "1.5"
+        os.environ["HINDSIGHT_API_REFLECT_LLM_MAX_BACKOFF"] = "90.0"
+
+        try:
+            clear_config_cache()
+            from hindsight_api.config import get_config
+
+            config = get_config()
+
+            # Verify retain overrides
+            assert config.retain_llm_max_retries == 3
+            assert config.retain_llm_initial_backoff == 2.0
+            assert config.retain_llm_max_backoff == 120.0
+
+            # Verify reflect overrides
+            assert config.reflect_llm_max_retries == 5
+            assert config.reflect_llm_initial_backoff == 1.5
+            assert config.reflect_llm_max_backoff == 90.0
+
+            # Verify global defaults remain unchanged
+            assert config.llm_max_retries == 10
+            assert config.llm_initial_backoff == 1.0
+            assert config.llm_max_backoff == 60.0
+        finally:
+            # Clean up
+            os.environ.pop("HINDSIGHT_API_RETAIN_LLM_MAX_RETRIES", None)
+            os.environ.pop("HINDSIGHT_API_RETAIN_LLM_INITIAL_BACKOFF", None)
+            os.environ.pop("HINDSIGHT_API_RETAIN_LLM_MAX_BACKOFF", None)
+            os.environ.pop("HINDSIGHT_API_REFLECT_LLM_MAX_RETRIES", None)
+            os.environ.pop("HINDSIGHT_API_REFLECT_LLM_INITIAL_BACKOFF", None)
+            os.environ.pop("HINDSIGHT_API_REFLECT_LLM_MAX_BACKOFF", None)
+            clear_config_cache()
+
+    def test_per_operation_retry_backoff_fallback_to_global(self):
+        """Test that per-operation settings fall back to global when not set."""
+        from hindsight_api.config import clear_config_cache, get_config
+
+        # Set only global values
+        os.environ["HINDSIGHT_API_LLM_MAX_RETRIES"] = "7"
+        os.environ["HINDSIGHT_API_LLM_INITIAL_BACKOFF"] = "3.0"
+        os.environ["HINDSIGHT_API_LLM_MAX_BACKOFF"] = "180.0"
+
+        try:
+            clear_config_cache()
+            config = get_config()
+
+            # Per-operation should be None (will fall back to global at runtime)
+            assert config.retain_llm_max_retries is None
+            assert config.retain_llm_initial_backoff is None
+            assert config.retain_llm_max_backoff is None
+
+            # Global values should be set
+            assert config.llm_max_retries == 7
+            assert config.llm_initial_backoff == 3.0
+            assert config.llm_max_backoff == 180.0
+        finally:
+            os.environ.pop("HINDSIGHT_API_LLM_MAX_RETRIES", None)
+            os.environ.pop("HINDSIGHT_API_LLM_INITIAL_BACKOFF", None)
+            os.environ.pop("HINDSIGHT_API_LLM_MAX_BACKOFF", None)
+            clear_config_cache()

hindsight-api/tests/test_reflect_agent.py

@@ -163,6 +163,12 @@ class TestToolNameNormalization:
         assert _normalize_tool_name("call=functions.recall") == "recall"
         assert _normalize_tool_name("call=functions.search_observations") == "search_observations"
 
+    def test_normalize_special_token_suffix(self):
+        """Tool names with malformed special tokens should be normalized."""
+        assert _normalize_tool_name("done<|channel|>commentary") == "done"
+        assert _normalize_tool_name("recall<|endoftext|>") == "recall"
+        assert _normalize_tool_name("search_observations<|im_end|>extra") == "search_observations"
+
     def test_is_done_tool(self):
         """Test _is_done_tool helper."""
         # Standard
@@ -174,9 +180,14 @@ class TestToolNameNormalization:
         assert _is_done_tool("call=done") is True
         assert _is_done_tool("call=functions.done") is True
 
+        # With malformed special tokens
+        assert _is_done_tool("done<|channel|>commentary") is True
+        assert _is_done_tool("done<|endoftext|>") is True
+
         # Not done
         assert _is_done_tool("functions.recall") is False
         assert _is_done_tool("call=functions.recall") is False
+        assert _is_done_tool("recall<|channel|>done") is False
 
 
 class TestReflectAgentMocked:

hindsight-api/tests/test_retain.py

@@ -16,7 +16,6 @@ async def test_retain_with_chunks(memory, request_context):
     Test that retain function:
     1. Stores facts with associated chunks
     2. Recall returns chunk_id for each fact
-    3. Recall with include_entities=True also works (for compatibility)
     """
     bank_id = f"test_chunks_{datetime.now(timezone.utc).timestamp()}"
     document_id = "test_doc_123"
@@ -56,7 +55,6 @@ async def test_retain_with_chunks(memory, request_context):
             budget=Budget.LOW,
             max_tokens=500,
             fact_type=["world"],  # Search for world facts
-            include_entities=False,  # Disable entities for simpler test
             include_chunks=True,  # Enable chunks
             max_chunk_tokens=8192,
             request_context=request_context,
@@ -146,7 +144,6 @@ async def test_chunks_and_entities_follow_fact_order(memory, request_context):
             budget=Budget.MID,
             max_tokens=1000,
             fact_type=["world"],
-            include_entities=True,
             include_chunks=True,
             max_chunk_tokens=8192,
             request_context=request_context,

hindsight-api/tests/test_tei_cross_encoder.py

@@ -527,6 +527,7 @@ class TestRemoteTEICrossEncoderConfig:
         """Test creating encoder from environment variables."""
         import os
 
+        from hindsight_api.config import clear_config_cache
         from hindsight_api.engine.cross_encoder import create_cross_encoder_from_env
 
         with patch.dict(
@@ -538,6 +539,7 @@ class TestRemoteTEICrossEncoderConfig:
                 "HINDSIGHT_API_RERANKER_TEI_MAX_CONCURRENT": "16",
             },
         ):
+            clear_config_cache()  # Clear cache to pick up patched env vars
             encoder = create_cross_encoder_from_env()
 
             assert isinstance(encoder, RemoteTEICrossEncoder)
@@ -545,6 +547,8 @@ class TestRemoteTEICrossEncoderConfig:
             assert encoder.batch_size == 256
             assert encoder.max_concurrent == 16
 
+        clear_config_cache()  # Clear cache after test
+
 
 # ============================================================================
 # TEI Reranker Performance Benchmark Tests

hindsight-api/tests/test_think.py

@@ -1,5 +1,5 @@
 """
-Test think function for opinion generation and consistency.
+Test reflect (think) function.
 """
 import pytest
 from datetime import datetime, timezone
@@ -7,131 +7,6 @@ from hindsight_api.engine.memory_engine import Budget
 from hindsight_api import RequestContext
 
 
-@pytest.mark.asyncio
-async def test_think_opinion_consistency(memory, request_context):
-    """
-    Test that think function:
-    1. Generates an opinion
-    2. Stores the opinion in the database
-    3. Returns consistent response on subsequent calls with the same query
-    """
-    bank_id = f"test_think_{datetime.now(timezone.utc).timestamp()}"
-
-    try:
-
-        # Store some initial facts to give context for opinion formation
-        await memory.retain_async(
-            bank_id=bank_id,
-            content="Alice is a software engineer who has worked on 5 major projects. She always delivers on time and writes clean, well-documented code.",
-            context="performance review",
-            event_date=datetime(2024, 1, 15, tzinfo=timezone.utc),
-            request_context=request_context,
-        )
-
-        await memory.retain_async(
-            bank_id=bank_id,
-            content="Bob recently joined the team. He missed his first deadline and his code had many bugs.",
-            context="performance review",
-            event_date=datetime(2024, 2, 1, tzinfo=timezone.utc),
-            request_context=request_context,
-        )
-
-        # First think call - should generate opinions
-        query = "Who is a more reliable engineer?"
-        result1 = await memory.reflect_async(
-            bank_id=bank_id,
-            query=query,
-            budget=Budget.LOW,
-            request_context=request_context,
-        )
-
-        print(f"\n=== First Think Call ===")
-        print(f"Answer: {result1.text}")
-
-        # Verify we got an answer
-        assert result1.text, "First think call should return an answer"
-        assert result1.based_on, "Should return based_on facts"
-
-        # Wait for background opinion processing tasks to complete
-        await memory.wait_for_background_tasks()
-
-        # Search for stored opinions to verify they were actually saved
-        pool = await memory._get_pool()
-        async with pool.acquire() as conn:
-            stored_opinions = await conn.fetch(
-                """
-                SELECT id, text, confidence_score, fact_type
-                FROM memory_units
-                WHERE bank_id = $1 AND fact_type = 'opinion'
-                ORDER BY created_at DESC
-                """,
-                bank_id
-            )
-
-        print(f"\n=== Stored Opinions in Database ===")
-        print(f"Total opinions stored: {len(stored_opinions)}")
-        for op in stored_opinions:
-            print(f"  - {op['text']} (confidence: {op['confidence_score']:.2f})")
-
-        # Verify opinions were actually written to database
-        # NOTE: Opinion extraction may not always detect opinions depending on the LLM response format
-        if len(stored_opinions) > 0:
-            assert all(op['fact_type'] == 'opinion' for op in stored_opinions), "All stored items should have fact_type='opinion'"
-            print(f"✓ Opinions were successfully stored in database")
-        else:
-            print(f"⚠ Note: No opinions were extracted/stored (this can happen if the LLM response format doesn't trigger opinion extraction)")
-
-        # Second think call - should use the stored opinions
-        result2 = await memory.reflect_async(
-            bank_id=bank_id,
-            query=query,
-            budget=Budget.LOW,
-            request_context=request_context,
-        )
-
-        print(f"\n=== Second Think Call ===")
-        print(f"Answer: {result2.text}")
-        print(f"Existing opinions used: {len(result2.based_on.get('opinion', []))}")
-        for opinion in result2.based_on.get('opinion', []):
-            print(f"  - {opinion.text}")
-
-        # Verify second call also got an answer
-        assert result2.text, "Second think call should return an answer"
-
-        # Verify second call used the stored opinions (if any were stored)
-        if len(stored_opinions) > 0:
-            assert len(result2.based_on.get('opinion', [])) > 0, "Second call should retrieve stored opinions"
-
-        # The responses should be consistent (both should mention the same person as more reliable)
-        # We'll do a basic check that they're not contradictory
-        text1_lower = result1.text.lower()
-        text2_lower = result2.text.lower()
-
-        print(f"\n=== Consistency Check ===")
-
-        # Check if Alice is mentioned as more reliable in first response
-        if 'alice' in text1_lower and ('reliable' in text1_lower or 'better' in text1_lower):
-            print("First response favors Alice")
-            # Second response should also favor Alice (consistency)
-            assert 'alice' in text2_lower, "Second response should also mention Alice"
-            print("Second response also mentions Alice - CONSISTENT ✓")
-
-        # Check if Bob is mentioned
-        if 'bob' in text1_lower:
-            print("First response mentions Bob")
-            if 'bob' in text2_lower:
-                print("Second response also mentions Bob - CONSISTENT ✓")
-
-        print(f"\n✅ Test passed - opinions were formed, stored, and used consistently")
-
-    finally:
-        # Clean up agent data
-        try:
-            await memory.delete_bank(bank_id, request_context=request_context)
-        except Exception as e:
-            print(f"Warning: Error during cleanup: {e}")
-
-
 @pytest.mark.asyncio
 async def test_think_without_prior_context(memory, request_context):
     """

hindsight-api/tests/test_vertexai_provider.py

@@ -0,0 +1,244 @@
+"""
+Test Vertex AI provider integration using native genai SDK.
+"""
+
+import os
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+# Skip all tests if google-auth not available
+pytest.importorskip("google.auth")
+
+
+def test_llm_wrapper_vertexai_missing_dependency():
+    """Test error when google-auth is not available and service account key is set."""
+    from hindsight_api.engine import llm_wrapper
+
+    # VERTEXAI_AVAILABLE only matters when a service account key is provided
+    original_available = llm_wrapper.VERTEXAI_AVAILABLE
+    try:
+        llm_wrapper.VERTEXAI_AVAILABLE = False
+
+        with patch.dict(
+            os.environ,
+            {
+                "HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID": "test-project",
+                "HINDSIGHT_API_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY": "/path/to/key.json",
+            },
+            clear=False,
+        ):
+            from hindsight_api.config import clear_config_cache
+
+            clear_config_cache()
+
+            with pytest.raises(ValueError, match="google-auth"):
+                from hindsight_api.engine.llm_wrapper import LLMProvider
+
+                LLMProvider(
+                    provider="vertexai",
+                    api_key="",
+                    base_url="",
+                    model="google/gemini-2.0-flash-001",
+                )
+
+            clear_config_cache()
+    finally:
+        llm_wrapper.VERTEXAI_AVAILABLE = original_available
+
+
+def test_llm_wrapper_vertexai_missing_project_id():
+    """Test error when project ID is not configured."""
+    with patch.dict(os.environ, {"HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID": ""}, clear=False):
+        from hindsight_api.config import clear_config_cache
+
+        clear_config_cache()
+
+        with pytest.raises(ValueError, match="HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID"):
+            from hindsight_api.engine.llm_wrapper import LLMProvider
+
+            LLMProvider(
+                provider="vertexai",
+                api_key="",
+                base_url="",
+                model="google/gemini-2.0-flash-001",
+            )
+
+        clear_config_cache()
+
+
+def test_llm_wrapper_vertexai_adc_auth():
+    """Test Vertex AI with ADC authentication creates native genai client."""
+    from hindsight_api.engine.llm_wrapper import LLMProvider
+
+    with patch.dict(
+        os.environ,
+        {"HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID": "test-project"},
+        clear=False,
+    ):
+        from hindsight_api.config import clear_config_cache
+
+        clear_config_cache()
+
+        # genai.Client handles ADC internally — just verify it creates the client
+        with patch("google.genai.Client") as mock_client_cls:
+            mock_client_cls.return_value = MagicMock()
+
+            provider = LLMProvider(
+                provider="vertexai",
+                api_key="",
+                base_url="",
+                model="google/gemini-2.0-flash-001",
+            )
+
+            assert provider.provider == "vertexai"
+            assert provider.model == "gemini-2.0-flash-001"  # google/ prefix stripped
+            assert provider._gemini_client is not None
+
+            # Verify genai.Client was called with vertexai=True
+            mock_client_cls.assert_called_once_with(
+                vertexai=True,
+                project="test-project",
+                location="us-central1",
+            )
+
+        clear_config_cache()
+
+
+def test_llm_wrapper_vertexai_sa_auth():
+    """Test Vertex AI with service account authentication passes credentials to genai client."""
+    from hindsight_api.engine.llm_wrapper import LLMProvider
+
+    mock_credentials = MagicMock()
+
+    with patch.dict(
+        os.environ,
+        {
+            "HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID": "test-project",
+            "HINDSIGHT_API_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY": "/path/to/key.json",
+        },
+        clear=False,
+    ):
+        from hindsight_api.config import clear_config_cache
+
+        clear_config_cache()
+
+        with patch(
+            "google.oauth2.service_account.Credentials.from_service_account_file",
+            return_value=mock_credentials,
+        ):
+            with patch("google.genai.Client") as mock_client_cls:
+                mock_client_cls.return_value = MagicMock()
+
+                provider = LLMProvider(
+                    provider="vertexai",
+                    api_key="",
+                    base_url="",
+                    model="google/gemini-2.0-flash-001",
+                )
+
+                assert provider.provider == "vertexai"
+                assert provider._gemini_client is not None
+
+                # Verify credentials were passed to genai.Client
+                mock_client_cls.assert_called_once_with(
+                    vertexai=True,
+                    project="test-project",
+                    location="us-central1",
+                    credentials=mock_credentials,
+                )
+
+        clear_config_cache()
+
+
+def test_llm_wrapper_vertexai_strips_google_prefix():
+    """Test that google/ prefix is stripped from model name for native SDK."""
+    from hindsight_api.engine.llm_wrapper import LLMProvider
+
+    with patch.dict(
+        os.environ,
+        {"HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID": "test-project"},
+        clear=False,
+    ):
+        from hindsight_api.config import clear_config_cache
+
+        clear_config_cache()
+
+        with patch("google.genai.Client") as mock_client_cls:
+            mock_client_cls.return_value = MagicMock()
+
+            provider = LLMProvider(
+                provider="vertexai",
+                api_key="",
+                base_url="",
+                model="google/gemini-2.0-flash-lite-001",
+            )
+
+            assert provider.model == "gemini-2.0-flash-lite-001"
+
+        clear_config_cache()
+
+
+def test_llm_wrapper_vertexai_no_prefix_model():
+    """Test that model without google/ prefix is unchanged."""
+    from hindsight_api.engine.llm_wrapper import LLMProvider
+
+    with patch.dict(
+        os.environ,
+        {"HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID": "test-project"},
+        clear=False,
+    ):
+        from hindsight_api.config import clear_config_cache
+
+        clear_config_cache()
+
+        with patch("google.genai.Client") as mock_client_cls:
+            mock_client_cls.return_value = MagicMock()
+
+            provider = LLMProvider(
+                provider="vertexai",
+                api_key="",
+                base_url="",
+                model="gemini-2.0-flash-001",
+            )
+
+            assert provider.model == "gemini-2.0-flash-001"
+
+        clear_config_cache()
+
+
+@pytest.mark.asyncio
+@pytest.mark.skipif(
+    not os.getenv("HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID"),
+    reason="Vertex AI integration tests require HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID",
+)
+async def test_vertexai_integration_actual_api():
+    """
+    Integration test with actual Vertex AI API.
+
+    Requires:
+    - HINDSIGHT_API_LLM_VERTEXAI_PROJECT_ID
+    - ADC or HINDSIGHT_API_LLM_VERTEXAI_SERVICE_ACCOUNT_KEY
+    """
+    from hindsight_api.engine.llm_wrapper import LLMProvider
+
+    provider = LLMProvider(
+        provider="vertexai",
+        api_key="",
+        base_url="",
+        model="google/gemini-2.0-flash-001",
+    )
+
+    try:
+        # Simple test call
+        response = await provider.call(
+            messages=[{"role": "user", "content": "Say 'ok' and nothing else"}],
+            max_completion_tokens=10,
+        )
+
+        assert response is not None
+        assert isinstance(response, str)
+        assert len(response) > 0
+
+    finally:
+        await provider.cleanup()

hindsight-api/tests/test_worker.py

@@ -156,7 +156,6 @@ class TestWorkerPoller:
             pool=pool,
             worker_id="test-worker-1",
             executor=mock_executor,
-            batch_size=10,
         )
 
         claimed = await poller.claim_batch()
@@ -177,8 +176,8 @@ class TestWorkerPoller:
             assert row["worker_id"] == "test-worker-1"
 
     @pytest.mark.asyncio
-    async def test_claim_batch_respects_batch_size(self, pool, clean_operations):
-        """Test that claim_batch respects the batch_size limit."""
+    async def test_claim_batch_respects_max_slots(self, pool, clean_operations):
+        """Test that claim_batch respects the max_slots limit."""
         from hindsight_api.worker import WorkerPoller
 
         # Create 10 pending tasks
@@ -196,12 +195,11 @@ class TestWorkerPoller:
                 payload,
             )
 
-        # Claim with batch_size=3
         poller = WorkerPoller(
             pool=pool,
             worker_id="test-worker-1",
             executor=lambda x: None,
-            batch_size=3,
+            max_slots=3,  # Limit to 3 concurrent tasks
         )
 
         claimed = await poller.claim_batch()
@@ -238,11 +236,14 @@ class TestWorkerPoller:
             executor=mock_executor,
         )
 
-        # Execute the task
+        # Execute the task (fire-and-forget)
         task_dict = json.loads(payload)
         claimed_task = ClaimedTask(operation_id=str(op_id), task_dict=task_dict, schema=None)
         await poller.execute_task(claimed_task)
 
+        # Wait for background task to complete
+        completed = await poller.wait_for_active_tasks(timeout=5.0)
+        assert completed, "Task did not complete within timeout"
         assert len(executed) == 1
 
         # Verify task is marked as completed
@@ -283,11 +284,15 @@ class TestWorkerPoller:
             max_retries=3,
         )
 
-        # Execute (should fail and retry)
+        # Execute (should fail and retry) - fire-and-forget
         task_dict = json.loads(payload)
         claimed_task = ClaimedTask(operation_id=str(op_id), task_dict=task_dict, schema=None)
         await poller.execute_task(claimed_task)
 
+        # Wait for background task to complete
+        completed = await poller.wait_for_active_tasks(timeout=5.0)
+        assert completed, "Task did not complete within timeout"
+
         # Verify task is back to pending with incremented retry_count
         row = await pool.fetchrow(
             "SELECT status, retry_count, worker_id FROM async_operations WHERE operation_id = $1",
@@ -327,11 +332,15 @@ class TestWorkerPoller:
             max_retries=3,
         )
 
-        # Execute (should fail permanently)
+        # Execute (should fail permanently) - fire-and-forget
         task_dict = json.loads(payload)
         claimed_task = ClaimedTask(operation_id=str(op_id), task_dict=task_dict, schema=None)
         await poller.execute_task(claimed_task)
 
+        # Wait for background task to complete
+        completed = await poller.wait_for_active_tasks(timeout=5.0)
+        assert completed, "Task did not complete within timeout"
+
         # Verify task is marked as failed
         row = await pool.fetchrow(
             "SELECT status, error_message FROM async_operations WHERE operation_id = $1",
@@ -388,7 +397,6 @@ class TestWorkerPoller:
             pool=pool,
             worker_id="test-worker-1",
             executor=lambda x: None,
-            batch_size=10,
         )
 
         claimed = await poller.claim_batch()
@@ -440,7 +448,6 @@ class TestWorkerPoller:
             pool=pool,
             worker_id="test-worker-1",
             executor=lambda x: None,
-            batch_size=10,
         )
 
         claimed = await poller.claim_batch()
@@ -607,7 +614,6 @@ class TestConcurrentWorkers:
                 pool=pool,
                 worker_id=worker_id,
                 executor=lambda x: None,
-                batch_size=5,  # Each worker tries to claim 5
             )
             claimed = await poller.claim_batch()
             workers_claimed[worker_id] = [task.operation_id for task in claimed]
@@ -680,7 +686,6 @@ class TestConcurrentWorkers:
             pool=pool,
             worker_id="new-worker",
             executor=lambda x: None,
-            batch_size=10,
         )
 
         claimed = await poller.claim_batch()
@@ -879,7 +884,6 @@ class TestDynamicTenantDiscovery:
             pool=pool,
             worker_id="test-worker-1",
             executor=lambda x: None,
-            batch_size=10,
             tenant_extension=mock_extension,
         )
 
@@ -946,7 +950,6 @@ class TestDynamicTenantDiscovery:
             pool=pool,
             worker_id="test-worker-1",
             executor=lambda x: None,
-            batch_size=10,
             tenant_extension=dynamic_extension,
         )
 
@@ -1008,7 +1011,6 @@ class TestDynamicTenantDiscovery:
             pool=pool,
             worker_id="test-worker-1",
             executor=lambda x: None,
-            batch_size=10,
         )
 
         claimed = await poller.claim_batch()
@@ -1017,3 +1019,198 @@ class TestDynamicTenantDiscovery:
         # All tasks should have schema=None (public)
         for task in claimed:
             assert task.schema is None
+
+
+async def test_worker_fire_and_forget_nonblocking(pool, clean_operations):
+    """
+    Test that worker continues polling while tasks run (fire-and-forget pattern).
+
+    This test verifies the FIX: With the old blocking behavior, the worker would
+    wait for all tasks in a batch to complete before claiming more. This test
+    would FAIL with the old code because tasks 3-4 wouldn't be claimed until
+    tasks 1-2 complete. With fire-and-forget, tasks 3-4 are claimed immediately.
+    """
+    from hindsight_api.worker.poller import WorkerPoller
+
+    task_started = {}  # operation_id -> Event (set when task starts)
+    task_canfinish = {}  # operation_id -> Event (wait before finishing)
+
+    async def blocking_executor(task_dict: dict):
+        op_id = task_dict["operation_id"]
+        # Signal that this task has started
+        started = asyncio.Event()
+        task_started[op_id] = started
+        started.set()
+
+        # Block until we're told to finish
+        finish = asyncio.Event()
+        task_canfinish[op_id] = finish
+        await finish.wait()
+
+    poller = WorkerPoller(
+        pool=pool,
+        worker_id="test-worker",
+        executor=blocking_executor,
+        poll_interval_ms=50,  # Fast polling
+        max_slots=10,
+        consolidation_max_slots=2,
+    )
+
+    bank_id = f"test-worker-{uuid.uuid4().hex[:8]}"
+
+    # Submit initial 2 tasks
+    task_ids = []
+    for i in range(2):
+        op_id = uuid.uuid4()
+        task_ids.append(str(op_id))
+        payload = json.dumps({"type": "test", "operation_type": "retain", "operation_id": str(op_id), "bank_id": bank_id})
+        await pool.execute(
+            """
+            INSERT INTO async_operations (operation_id, bank_id, operation_type, status, task_payload)
+            VALUES ($1, $2, 'retain', 'pending', $3::jsonb)
+            """,
+            op_id,
+            bank_id,
+            payload,
+        )
+
+    poll_task = asyncio.create_task(poller.run())
+
+    try:
+        # Wait for first 2 tasks to start executing (but not finish)
+        for i in range(100):  # Try for up to 1 second
+            if len(task_started) >= 2:
+                break
+            await asyncio.sleep(0.01)
+        assert len(task_started) == 2, f"Expected 2 tasks started, got {len(task_started)}"
+
+        # Verify tasks are in_flight
+        async with poller._in_flight_lock:
+            assert poller._in_flight_count == 2
+
+        # NOW submit 2 more tasks WHILE the first 2 are still running
+        for i in range(2):
+            op_id = uuid.uuid4()
+            task_ids.append(str(op_id))
+            payload = json.dumps({"type": "test", "operation_type": "retain", "operation_id": str(op_id), "bank_id": bank_id})
+            await pool.execute(
+                """
+                INSERT INTO async_operations (operation_id, bank_id, operation_type, status, task_payload)
+                VALUES ($1, $2, 'retain', 'pending', $3::jsonb)
+                """,
+                op_id,
+                bank_id,
+                payload,
+            )
+
+        # KEY ASSERTION: Worker should claim tasks 3-4 WITHOUT waiting for 1-2 to finish
+        # This would FAIL with the old blocking behavior
+        for i in range(100):  # Try for up to 1 second
+            if len(task_started) >= 4:
+                break
+            await asyncio.sleep(0.01)
+
+        assert len(task_started) == 4, (
+            f"Fire-and-forget FAILED: Expected 4 tasks started, got {len(task_started)}. "
+            "This means the worker blocked waiting for the first batch to complete."
+        )
+
+        # Verify all 4 tasks are in-flight
+        async with poller._in_flight_lock:
+            assert poller._in_flight_count == 4
+
+        # Clean up: allow all tasks to finish
+        for event in task_canfinish.values():
+            event.set()
+
+    finally:
+        # Ensure cleanup
+        for event in task_canfinish.values():
+            event.set()
+        await poller.shutdown_graceful(timeout=2.0)
+        try:
+            await asyncio.wait_for(poll_task, timeout=1.0)
+        except asyncio.CancelledError:
+            pass
+
+
+async def test_worker_slot_limits_enforced(pool, clean_operations):
+    """Test that worker respects max_slots and won't exceed the limit."""
+    from hindsight_api.worker.poller import WorkerPoller
+
+    tasks_started = set()
+    task_events = {}
+
+    async def controlled_executor(task_dict: dict):
+        op_id = task_dict["operation_id"]
+        tasks_started.add(op_id)
+        event = asyncio.Event()
+        task_events[op_id] = event
+        await event.wait()
+
+    poller = WorkerPoller(
+        pool=pool,
+        worker_id="test-worker",
+        executor=controlled_executor,
+        poll_interval_ms=50,
+        max_slots=3,  # Only allow 3 concurrent tasks
+        consolidation_max_slots=1,
+    )
+
+    # Submit 10 tasks
+    bank_id = f"test-worker-{uuid.uuid4().hex[:8]}"
+    for i in range(10):
+        op_id = uuid.uuid4()
+        payload = json.dumps({"type": "test", "operation_type": "retain", "operation_id": str(op_id), "bank_id": bank_id})
+        await pool.execute(
+            """
+            INSERT INTO async_operations (operation_id, bank_id, operation_type, status, task_payload)
+            VALUES ($1, $2, 'retain', 'pending', $3::jsonb)
+            """,
+            op_id,
+            bank_id,
+            payload,
+        )
+
+    poll_task = asyncio.create_task(poller.run())
+
+    try:
+        # Wait for slots to fill
+        for i in range(100):
+            if len(tasks_started) >= 3:
+                break
+            await asyncio.sleep(0.01)
+
+        # Should have claimed exactly 3 tasks (slot limit)
+        assert len(tasks_started) == 3
+
+        # Wait to ensure no additional tasks are claimed
+        for i in range(30):
+            await asyncio.sleep(0.01)
+        assert len(tasks_started) == 3, "Worker exceeded slot limit!"
+
+        # Release tasks one by one and verify remaining are claimed
+        completed = 0
+        while completed < 10 and len(tasks_started) < 10:
+            # Release the next batch
+            events_to_release = list(task_events.values())[completed:completed+3]
+            for event in events_to_release:
+                event.set()
+            completed += len(events_to_release)
+
+            # Wait for new tasks to be claimed
+            for i in range(100):
+                if len(tasks_started) >= min(completed + 3, 10):
+                    break
+                await asyncio.sleep(0.01)
+
+        assert len(tasks_started) == 10
+
+    finally:
+        for event in task_events.values():
+            event.set()
+        await poller.shutdown_graceful(timeout=2.0)
+        try:
+            await asyncio.wait_for(poll_task, timeout=1.0)
+        except asyncio.CancelledError:
+            pass

hindsight-cli/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "hindsight-cli"
-version = "0.3.0"
+version = "0.4.2"
 edition = "2021"
 authors = ["Hindsight Team"]
 description = "A beautiful CLI for Hindsight - semantic memory system"

hindsight-cli/src/commands/bank.rs

@@ -500,6 +500,8 @@ pub fn delete(
 pub fn consolidate(
     client: &ApiClient,
     bank_id: &str,
+    wait: bool,
+    poll_interval: u64,
     verbose: bool,
     output_format: OutputFormat,
 ) -> Result<()> {
@@ -517,17 +519,82 @@ pub fn consolidate(
 
     match response {
         Ok(result) => {
+            let operation_id = result.operation_id.clone();
+
             if output_format == OutputFormat::Pretty {
                 ui::print_success("Consolidation triggered");
-                println!("  {} {}", ui::dim("Operation ID:"), result.operation_id);
+                println!("  {} {}", ui::dim("Operation ID:"), operation_id);
                 if result.deduplicated {
                     println!("  {} {}", ui::dim("Note:"), "Reusing existing pending consolidation task");
                 }
-                println!();
-                println!("{}", ui::dim("Use 'hindsight operation get' to check the operation status."));
             } else {
                 output::print_output(&result, output_format)?;
             }
+
+            if !wait {
+                if output_format == OutputFormat::Pretty {
+                    println!();
+                    println!("{}", ui::dim("Use --wait to poll for completion, or 'hindsight operation get' to check status."));
+                }
+                return Ok(());
+            }
+
+            // Poll for completion
+            if output_format == OutputFormat::Pretty {
+                println!();
+                println!("{}", ui::dim(&format!("Polling every {}s for completion...", poll_interval)));
+            }
+
+            let start = std::time::Instant::now();
+            loop {
+                std::thread::sleep(std::time::Duration::from_secs(poll_interval));
+                let elapsed = start.elapsed().as_secs();
+
+                let ops_result = client.list_operations(bank_id, verbose);
+                match ops_result {
+                    Ok(ops) => {
+                        // Find the operation by ID
+                        let op = ops.operations.iter().find(|o| o.id == operation_id);
+
+                        match op.map(|o| o.status.as_str()) {
+                            Some("completed") => {
+                                if output_format == OutputFormat::Pretty {
+                                    ui::print_success(&format!("Consolidation completed ({}s)", elapsed));
+                                }
+                                break;
+                            }
+                            Some("failed") => {
+                                let error_msg = op
+                                    .and_then(|o| o.error_message.as_ref())
+                                    .map(|s| s.as_str())
+                                    .unwrap_or("Unknown error");
+                                if output_format == OutputFormat::Pretty {
+                                    ui::print_error(&format!("Consolidation failed: {}", error_msg));
+                                }
+                                std::process::exit(1);
+                            }
+                            Some(status) => {
+                                if output_format == OutputFormat::Pretty {
+                                    println!("  ⏳ {} ({}s elapsed)", status, elapsed);
+                                }
+                            }
+                            None => {
+                                if output_format == OutputFormat::Pretty {
+                                    ui::print_warning(&format!("Operation {} not found in list", operation_id));
+                                }
+                                break;
+                            }
+                        }
+                    }
+                    Err(e) => {
+                        if output_format == OutputFormat::Pretty {
+                            ui::print_error(&format!("Failed to check operation status: {}", e));
+                        }
+                        return Err(e);
+                    }
+                }
+            }
+
             Ok(())
         }
         Err(e) => Err(e),

hindsight-cli/src/commands/document.rs

@@ -1,4 +1,6 @@
 use anyhow::Result;
+use chrono::{Duration as ChronoDuration, NaiveDate, Utc};
+use std::collections::BTreeMap;
 use crate::api::ApiClient;
 use crate::output::{self, OutputFormat};
 use crate::ui;
@@ -7,11 +9,17 @@ pub fn list(
     client: &ApiClient,
     agent_id: &str,
     query: Option<String>,
+    date: Option<String>,
     limit: i32,
     offset: i32,
     verbose: bool,
     output_format: OutputFormat,
 ) -> Result<()> {
+    // If date filter is provided, use the date-aware listing
+    if date.is_some() {
+        return list_with_date(client, agent_id, date.as_deref(), verbose, output_format);
+    }
+
     let spinner = if output_format == OutputFormat::Pretty {
         Some(ui::create_spinner("Fetching documents..."))
     } else {
@@ -50,6 +58,139 @@ pub fn list(
     }
 }
 
+/// List documents with date filtering
+fn list_with_date(
+    client: &ApiClient,
+    bank_id: &str,
+    date_filter: Option<&str>,
+    verbose: bool,
+    output_format: OutputFormat,
+) -> Result<()> {
+    let spinner = if output_format == OutputFormat::Pretty {
+        Some(ui::create_spinner("Fetching all documents..."))
+    } else {
+        None
+    };
+
+    // Fetch all documents with pagination
+    let all_docs = fetch_all_documents(client, bank_id, verbose)?;
+
+    if let Some(mut sp) = spinner {
+        sp.finish();
+    }
+
+    // Parse the date filter
+    let target_date = parse_date_filter(date_filter)?;
+
+    // Filter and group documents by date
+    let mut by_date: BTreeMap<String, Vec<serde_json::Value>> = BTreeMap::new();
+    let mut filtered_count = 0;
+
+    for doc in all_docs {
+        let created_at = doc.get("created_at")
+            .and_then(|v| v.as_str())
+            .unwrap_or("");
+
+        // Parse the date part (YYYY-MM-DD) from created_at
+        let doc_date = created_at.split('T').next().unwrap_or("");
+
+        // Apply date filter if specified
+        if let Some(ref target) = target_date {
+            let target_str = target.format("%Y-%m-%d").to_string();
+            if doc_date != target_str {
+                continue;
+            }
+        }
+
+        filtered_count += 1;
+        by_date.entry(doc_date.to_string()).or_default().push(doc);
+    }
+
+    // Output
+    if output_format == OutputFormat::Pretty {
+        let filter_desc = match date_filter {
+            None | Some("yesterday") => "yesterday".to_string(),
+            Some("today") => "today".to_string(),
+            Some("all") => "all dates".to_string(),
+            Some(d) => d.to_string(),
+        };
+
+        ui::print_info(&format!(
+            "Documents for bank '{}' (filter: {}, showing: {})",
+            bank_id, filter_desc, filtered_count
+        ));
+        println!();
+
+        // Show documents grouped by date (reverse order - newest first)
+        for (date_str, docs) in by_date.iter().rev() {
+            println!("  {} ({} documents)", date_str, docs.len());
+            for doc in docs {
+                let id = doc.get("id").and_then(|v| v.as_str()).unwrap_or("unknown");
+                let mem_count = doc.get("memory_unit_count").and_then(|v| v.as_i64()).unwrap_or(0);
+                println!("    - {} ({} memories)", id, mem_count);
+            }
+            println!();
+        }
+    } else {
+        // JSON/YAML output - convert to a list structure
+        let output: Vec<serde_json::Value> = by_date.values().flatten().cloned().collect();
+        output::print_output(&output, output_format)?;
+    }
+
+    Ok(())
+}
+
+/// Fetch all documents with pagination
+fn fetch_all_documents(
+    client: &ApiClient,
+    bank_id: &str,
+    verbose: bool,
+) -> Result<Vec<serde_json::Value>> {
+    let mut all_docs = Vec::new();
+    let mut offset = 0;
+    let limit = 500;
+
+    loop {
+        let response = client.list_documents(bank_id, None, Some(limit), Some(offset), verbose)?;
+
+        if response.items.is_empty() {
+            break;
+        }
+
+        // Convert Map<String, Value> to Value for each item
+        for item in response.items {
+            all_docs.push(serde_json::Value::Object(item));
+        }
+
+        offset += limit;
+
+        // Check if we've fetched everything
+        if all_docs.len() >= response.total as usize {
+            break;
+        }
+    }
+
+    Ok(all_docs)
+}
+
+/// Parse date filter string into a NaiveDate
+fn parse_date_filter(filter: Option<&str>) -> Result<Option<NaiveDate>> {
+    match filter {
+        None | Some("yesterday") => {
+            // Default to yesterday
+            Ok(Some(Utc::now().date_naive() - ChronoDuration::days(1)))
+        }
+        Some("today") => Ok(Some(Utc::now().date_naive())),
+        Some("all") => Ok(None), // No filtering
+        Some(date_str) => {
+            // Try to parse as YYYY-MM-DD
+            NaiveDate::parse_from_str(date_str, "%Y-%m-%d")
+                .map(Some)
+                .map_err(|e| anyhow::anyhow!("Invalid date format '{}': {}. Use YYYY-MM-DD, 'yesterday', 'today', or 'all'", date_str, e))
+        }
+    }
+}
+
 pub fn get(
     client: &ApiClient,
     agent_id: &str,

hindsight-cli/src/main.rs

@@ -260,6 +260,14 @@ enum BankCommands {
     Consolidate {
         /// Bank ID
         bank_id: String,
+
+        /// Wait for consolidation to complete (poll for status)
+        #[arg(long)]
+        wait: bool,
+
+        /// Poll interval in seconds (only used with --wait)
+        #[arg(long, default_value = "10")]
+        poll_interval: u64,
     },
 
     /// Clear all observations for a bank
@@ -441,6 +449,10 @@ enum DocumentCommands {
         #[arg(short = 'q', long)]
         query: Option<String>,
 
+        /// Filter by date (yesterday, today, YYYY-MM-DD, or all)
+        #[arg(short = 'd', long)]
+        date: Option<String>,
+
         /// Maximum number of results
         #[arg(short = 'l', long, default_value = "100")]
         limit: i32,
@@ -754,8 +766,8 @@ fn run() -> Result<()> {
             BankCommands::Delete { bank_id, yes } => {
                 commands::bank::delete(&client, &bank_id, yes, verbose, output_format)
             }
-            BankCommands::Consolidate { bank_id } => {
-                commands::bank::consolidate(&client, &bank_id, verbose, output_format)
+            BankCommands::Consolidate { bank_id, wait, poll_interval } => {
+                commands::bank::consolidate(&client, &bank_id, wait, poll_interval, verbose, output_format)
             }
             BankCommands::ClearObservations { bank_id, yes } => {
                 commands::bank::clear_observations(&client, &bank_id, yes, verbose, output_format)
@@ -792,8 +804,8 @@ fn run() -> Result<()> {
 
         // Document commands
         Commands::Document(doc_cmd) => match doc_cmd {
-            DocumentCommands::List { bank_id, query, limit, offset } => {
-                commands::document::list(&client, &bank_id, query, limit, offset, verbose, output_format)
+            DocumentCommands::List { bank_id, query, date, limit, offset } => {
+                commands::document::list(&client, &bank_id, query, date, limit, offset, verbose, output_format)
             }
             DocumentCommands::Get { bank_id, document_id } => {
                 commands::document::get(&client, &bank_id, &document_id, verbose, output_format)

hindsight-cli/tests/integration_test.rs

@@ -0,0 +1,125 @@
+use std::process::Command;
+
+#[test]
+fn test_cli_help() {
+    let output = Command::new("cargo")
+        .args(["run", "--", "--help"])
+        .output()
+        .expect("Failed to execute command");
+
+    assert!(output.status.success());
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    assert!(stdout.contains("Hindsight CLI"));
+}
+
+#[test]
+fn test_cli_version() {
+    let output = Command::new("cargo")
+        .args(["run", "--", "--version"])
+        .output()
+        .expect("Failed to execute command");
+
+    assert!(output.status.success());
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    assert!(stdout.contains("hindsight"));
+}
+
+#[test]
+fn test_ui_command_without_config() {
+    // Test that the ui command handles missing config gracefully
+    // Create a temp home directory with no config
+    let temp_dir = std::env::temp_dir().join(format!("hindsight-test-ui-{}", std::process::id()));
+    std::fs::create_dir_all(&temp_dir).expect("Failed to create temp dir");
+
+    let output = Command::new("cargo")
+        .args(["run", "--", "ui"])
+        .env_remove("HINDSIGHT_API_URL")
+        .env_remove("HINDSIGHT_API_KEY")
+        .env("HOME", &temp_dir)
+        .output()
+        .expect("Failed to execute command");
+
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    let stderr = String::from_utf8_lossy(&output.stderr);
+
+    // Either it fails with a config error or it succeeds if there's a default config
+    // Just verify it doesn't crash unexpectedly
+    assert!(
+        !output.status.success()
+        || stdout.contains("Launching Hindsight Control Plane UI")
+        || stderr.contains("Configuration error")
+        || stderr.contains("HINDSIGHT_API_URL"),
+        "Unexpected output - stdout: {}, stderr: {}",
+        stdout,
+        stderr
+    );
+
+    // Cleanup
+    std::fs::remove_dir_all(&temp_dir).ok();
+}
+
+#[test]
+fn test_ui_command_with_config() {
+    // This test is skipped by default since it requires a running control plane
+    // and would block for a long time. The other tests cover the basic functionality.
+    // To run this test manually:
+    // 1. Build the control plane: cd hindsight-control-plane && npm run build
+    // 2. Run: cargo test test_ui_command_with_config -- --ignored
+
+    // Just verify that the ui command accepts the configuration
+    let temp_dir = std::env::temp_dir().join(format!("hindsight-test-ui-valid-{}", std::process::id()));
+    std::fs::create_dir_all(&temp_dir).expect("Failed to create temp dir");
+
+    // Write a minimal config
+    let config_dir = temp_dir.join(".config").join("hindsight");
+    std::fs::create_dir_all(&config_dir).expect("Failed to create config dir");
+    let config_file = config_dir.join("config");
+    std::fs::write(&config_file, "api_url=http://localhost:8888\napi_key=test-key\n")
+        .expect("Failed to write config");
+
+    let output = Command::new("cargo")
+        .args(["run", "--", "ui", "--help"])
+        .env("HOME", &temp_dir)
+        .output()
+        .expect("Failed to execute command");
+
+    // The --help should work regardless
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    assert!(stdout.contains("Hindsight CLI") || output.status.success());
+
+    // Cleanup
+    std::fs::remove_dir_all(&temp_dir).ok();
+}
+
+#[test]
+fn test_configure_command() {
+    // Test that configure command creates/updates config
+    let temp_dir = std::env::temp_dir().join(format!("hindsight-test-{}", std::process::id()));
+    std::fs::create_dir_all(&temp_dir).expect("Failed to create temp dir");
+
+    let output = Command::new("cargo")
+        .args([
+            "run",
+            "--",
+            "configure",
+            "--api-url",
+            "http://localhost:9999",
+            "--api-key",
+            "test-key-123"
+        ])
+        .env("HOME", &temp_dir)
+        .output()
+        .expect("Failed to execute command");
+
+    assert!(
+        output.status.success(),
+        "Configure command failed: {}",
+        String::from_utf8_lossy(&output.stderr)
+    );
+
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    assert!(stdout.contains("Configuration saved") || stdout.contains("success"));
+
+    // Cleanup
+    std::fs::remove_dir_all(&temp_dir).ok();
+}

hindsight-clients/python/hindsight_client_api/__init__.py

@@ -7,7 +7,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/banks_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/directives_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/documents_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/entities_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/memory_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.
@@ -1644,7 +1644,7 @@ class MemoryApi:
     ) -> RecallResponse:
         """Recall memory
 
-        Recall memory using semantic similarity and spreading activation.  The type parameter is optional and must be one of: - `world`: General knowledge about people, places, events, and things that happen - `experience`: Memories about experience, conversations, actions taken, and tasks performed - `opinion`: The bank's formed beliefs, perspectives, and viewpoints  Set `include_entities=true` to get entity observations alongside recall results.
+        Recall memory using semantic similarity and spreading activation.  The type parameter is optional and must be one of: - `world`: General knowledge about people, places, events, and things that happen - `experience`: Memories about experience, conversations, actions taken, and tasks performed
 
         :param bank_id: (required)
         :type bank_id: str
@@ -1720,7 +1720,7 @@ class MemoryApi:
     ) -> ApiResponse[RecallResponse]:
         """Recall memory
 
-        Recall memory using semantic similarity and spreading activation.  The type parameter is optional and must be one of: - `world`: General knowledge about people, places, events, and things that happen - `experience`: Memories about experience, conversations, actions taken, and tasks performed - `opinion`: The bank's formed beliefs, perspectives, and viewpoints  Set `include_entities=true` to get entity observations alongside recall results.
+        Recall memory using semantic similarity and spreading activation.  The type parameter is optional and must be one of: - `world`: General knowledge about people, places, events, and things that happen - `experience`: Memories about experience, conversations, actions taken, and tasks performed
 
         :param bank_id: (required)
         :type bank_id: str
@@ -1796,7 +1796,7 @@ class MemoryApi:
     ) -> RESTResponseType:
         """Recall memory
 
-        Recall memory using semantic similarity and spreading activation.  The type parameter is optional and must be one of: - `world`: General knowledge about people, places, events, and things that happen - `experience`: Memories about experience, conversations, actions taken, and tasks performed - `opinion`: The bank's formed beliefs, perspectives, and viewpoints  Set `include_entities=true` to get entity observations alongside recall results.
+        Recall memory using semantic similarity and spreading activation.  The type parameter is optional and must be one of: - `world`: General knowledge about people, places, events, and things that happen - `experience`: Memories about experience, conversations, actions taken, and tasks performed
 
         :param bank_id: (required)
         :type bank_id: str
@@ -1950,7 +1950,7 @@ class MemoryApi:
     ) -> ReflectResponse:
         """Reflect and generate answer
 
-        Reflect and formulate an answer using bank identity, world facts, and opinions.  This endpoint: 1. Retrieves experience (conversations and events) 2. Retrieves world facts relevant to the query 3. Retrieves existing opinions (bank's perspectives) 4. Uses LLM to formulate a contextual answer 5. Extracts and stores any new opinions formed 6. Returns plain text answer, the facts used, and new opinions
+        Reflect and formulate an answer using bank identity, world facts, and opinions.  This endpoint: 1. Retrieves experience (conversations and events) 2. Retrieves world facts relevant to the query 3. Retrieves existing opinions (bank's perspectives) 4. Uses LLM to formulate a contextual answer 5. Returns plain text answer and the facts used
 
         :param bank_id: (required)
         :type bank_id: str
@@ -2026,7 +2026,7 @@ class MemoryApi:
     ) -> ApiResponse[ReflectResponse]:
         """Reflect and generate answer
 
-        Reflect and formulate an answer using bank identity, world facts, and opinions.  This endpoint: 1. Retrieves experience (conversations and events) 2. Retrieves world facts relevant to the query 3. Retrieves existing opinions (bank's perspectives) 4. Uses LLM to formulate a contextual answer 5. Extracts and stores any new opinions formed 6. Returns plain text answer, the facts used, and new opinions
+        Reflect and formulate an answer using bank identity, world facts, and opinions.  This endpoint: 1. Retrieves experience (conversations and events) 2. Retrieves world facts relevant to the query 3. Retrieves existing opinions (bank's perspectives) 4. Uses LLM to formulate a contextual answer 5. Returns plain text answer and the facts used
 
         :param bank_id: (required)
         :type bank_id: str
@@ -2102,7 +2102,7 @@ class MemoryApi:
     ) -> RESTResponseType:
         """Reflect and generate answer
 
-        Reflect and formulate an answer using bank identity, world facts, and opinions.  This endpoint: 1. Retrieves experience (conversations and events) 2. Retrieves world facts relevant to the query 3. Retrieves existing opinions (bank's perspectives) 4. Uses LLM to formulate a contextual answer 5. Extracts and stores any new opinions formed 6. Returns plain text answer, the facts used, and new opinions
+        Reflect and formulate an answer using bank identity, world facts, and opinions.  This endpoint: 1. Retrieves experience (conversations and events) 2. Retrieves world facts relevant to the query 3. Retrieves existing opinions (bank's perspectives) 4. Uses LLM to formulate a contextual answer 5. Returns plain text answer and the facts used
 
         :param bank_id: (required)
         :type bank_id: str

hindsight-clients/python/hindsight_client_api/api/mental_models_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/monitoring_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api/operations_api.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/api_client.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/configuration.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.
@@ -489,7 +489,7 @@ class Configuration:
         return "Python SDK Debug Report:\n"\
                "OS: {env}\n"\
                "Python Version: {pyversion}\n"\
-               "Version of the API: 0.1.0\n"\
+               "Version of the API: 0.4.2\n"\
                "SDK Package Version: 0.0.7".\
                format(env=sys.platform, pyversion=sys.version)
 

hindsight-clients/python/hindsight_client_api/exceptions.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/__init__.py

@@ -6,7 +6,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/add_background_request.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/async_operation_submit_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/background_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/bank_list_item.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/bank_list_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/bank_profile_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/bank_stats_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/budget.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/cancel_operation_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/chunk_data.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/chunk_include_options.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/chunk_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/consolidation_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/create_bank_request.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/create_directive_request.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/create_mental_model_request.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/create_mental_model_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/delete_document_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/delete_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/directive_list_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/directive_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/disposition_traits.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/document_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_detail_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_include_options.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_input.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_list_item.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_list_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_observation_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/entity_state_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/features_info.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/graph_data_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/http_validation_error.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/include_options.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.

hindsight-clients/python/hindsight_client_api/models/list_documents_response.py

@@ -5,7 +5,7 @@
 
     HTTP API for Hindsight
 
-    The version of the OpenAPI document: 0.1.0
+    The version of the OpenAPI document: 0.4.2
     Generated by OpenAPI Generator (https://openapi-generator.tech)
 
     Do not edit the class manually.