hindsight-api 0.4.1__py3-none-any.whl → 0.4.3__py3-none-any.whl

hindsight_api/engine/memory_engine.py

@@ -504,12 +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)
@@ -789,7 +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()
@@ -889,6 +888,23 @@ class MemoryEngine(MemoryEngineInterface):
             # 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, schema=get_config().database_schema)
@@ -1175,15 +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:
@@ -1231,8 +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:
@@ -1554,16 +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={})
@@ -2219,44 +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"),
@@ -2267,38 +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
 
@@ -2367,7 +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:
@@ -2376,7 +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))
@@ -3550,7 +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
@@ -3875,7 +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,
@@ -3904,32 +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,
@@ -4116,36 +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)
     # =========================================================================
@@ -4256,9 +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"],
@@ -4266,7 +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/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/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/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/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/main.py

@@ -140,13 +140,6 @@ def main():
         args.port = DEFAULT_DAEMON_PORT
         args.host = "127.0.0.1"  # Only bind to localhost for security
 
-        # Force CPU mode for daemon to avoid macOS MPS/XPC issues
-        # MPS (Metal Performance Shaders) has unstable XPC connections in background processes
-        # that can cause assertion failures and process crashes at the C++ level
-        # (which Python exception handlers cannot catch)
-        os.environ["HINDSIGHT_API_EMBEDDINGS_LOCAL_FORCE_CPU"] = "1"
-        os.environ["HINDSIGHT_API_RERANKER_LOCAL_FORCE_CPU"] = "1"
-
         # Check if another daemon is already running
         daemon_lock = DaemonLock()
         if not daemon_lock.acquire():
@@ -183,19 +176,40 @@ def main():
             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,
@@ -225,7 +239,6 @@ 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,
@@ -240,8 +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,
         )
@@ -353,6 +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():
@@ -363,12 +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__":