gnosisllm-knowledge 0.2.0__py3-none-any.whl → 0.4.0__py3-none-any.whl

gnosisllm_knowledge/backends/opensearch/memory/client.py

@@ -0,0 +1,1380 @@
+"""OpenSearch Agentic Memory API client."""
+
+from __future__ import annotations
+
+import logging
+from datetime import datetime
+from typing import TYPE_CHECKING, Any
+
+import httpx
+
+from gnosisllm_knowledge.core.domain.memory import (
+    ContainerConfig,
+    ContainerInfo,
+    HistoryAction,
+    HistoryEntry,
+    MemoryEntry,
+    MemoryStats,
+    MemoryStrategy,
+    MemoryType,
+    Message,
+    Namespace,
+    PayloadType,
+    RecallResult,
+    SessionInfo,
+    StoreRequest,
+    StoreResult,
+)
+from gnosisllm_knowledge.core.exceptions import ContainerNotFoundError, MemoryError
+
+if TYPE_CHECKING:
+    from gnosisllm_knowledge.backends.opensearch.memory.config import MemoryConfig
+
+logger = logging.getLogger(__name__)
+
+
+class OpenSearchMemoryClient:
+    """Client for OpenSearch Agentic Memory APIs.
+
+    Implements memory operations using the OpenSearch ML Memory plugin.
+
+    Example:
+        ```python
+        client = OpenSearchMemoryClient(config)
+
+        # Create container
+        container = await client.create_container(ContainerConfig(
+            name="my-memory",
+            strategies=[
+                StrategyConfig(type=MemoryStrategy.SEMANTIC, namespace=["user_id"]),
+            ],
+        ))
+
+        # List containers
+        containers = await client.list_containers()
+        ```
+    """
+
+    def __init__(self, config: MemoryConfig) -> None:
+        """Initialize the client.
+
+        Args:
+            config: Memory configuration.
+        """
+        self._config = config
+        self._base_url = config.url
+        self._auth = config.auth
+
+    # === Container Management ===
+
+    async def create_container(
+        self,
+        config: ContainerConfig,
+        **options: Any,
+    ) -> ContainerInfo:
+        """Create a memory container.
+
+        Args:
+            config: Container configuration.
+            **options: Additional options.
+
+        Returns:
+            Created container info.
+        """
+        # Build strategies list
+        strategies = [s.to_dict() for s in config.strategies]
+        if not strategies:
+            # Use default strategies from config
+            strategies = [
+                {
+                    "type": s.value,
+                    "namespace": ["user_id"],
+                    "configuration": {
+                        "llm_result_path": self._config.llm_result_path,
+                    },
+                }
+                for s in self._config.default_strategies
+            ]
+
+        body: dict[str, Any] = {
+            "name": config.name,
+            "configuration": {
+                "embedding_model_id": config.embedding_model_id
+                or self._config.embedding_model_id,
+                "embedding_model_type": config.embedding_model_type.value,
+                "embedding_dimension": config.embedding_dimension,
+                "llm_id": config.llm_model_id or self._config.llm_model_id,
+                "strategies": strategies,
+                "use_system_index": config.use_system_index,
+                "parameters": {
+                    "llm_result_path": config.llm_result_path,
+                },
+            },
+        }
+
+        if config.description:
+            body["description"] = config.description
+        if config.index_prefix:
+            body["configuration"]["index_prefix"] = config.index_prefix
+        if config.index_settings:
+            body["configuration"]["index_settings"] = config.index_settings.to_dict()
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/_create",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        container_id = result.get("memory_container_id")
+
+        return ContainerInfo(
+            id=container_id,
+            name=config.name,
+            description=config.description,
+            strategies=[s.type for s in config.strategies] if config.strategies else [],
+            embedding_model_id=config.embedding_model_id,
+            llm_model_id=config.llm_model_id,
+        )
+
+    async def get_container(
+        self,
+        container_id: str,
+        **options: Any,
+    ) -> ContainerInfo | None:
+        """Get container by ID.
+
+        Args:
+            container_id: Container ID.
+            **options: Additional options.
+
+        Returns:
+            Container info or None if not found.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.get(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return None
+            response.raise_for_status()
+            data = response.json()
+
+        return self._parse_container_info(container_id, data)
+
+    async def list_containers(
+        self,
+        limit: int = 100,
+        **options: Any,
+    ) -> list[ContainerInfo]:
+        """List all containers.
+
+        Args:
+            limit: Maximum number of containers to return.
+            **options: Additional options.
+
+        Returns:
+            List of container info.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/_search",
+                json={"query": {"match_all": {}}, "size": limit},
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            data = response.json()
+
+        containers = []
+        for hit in data.get("hits", {}).get("hits", []):
+            container = self._parse_container_info(hit["_id"], hit["_source"])
+            containers.append(container)
+
+        return containers
+
+    async def update_container(
+        self,
+        container_id: str,
+        config: ContainerConfig,
+        **options: Any,
+    ) -> ContainerInfo:
+        """Update a container configuration.
+
+        Args:
+            container_id: Container ID.
+            config: Updated configuration.
+            **options: Additional options.
+
+        Returns:
+            Updated container info.
+        """
+        body: dict[str, Any] = {}
+
+        if config.description is not None:
+            body["description"] = config.description
+
+        if config.strategies:
+            body["configuration"] = {
+                "strategies": [s.to_dict() for s in config.strategies],
+            }
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.put(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            data = response.json()
+
+        return self._parse_container_info(container_id, data)
+
+    async def delete_container(
+        self,
+        container_id: str,
+        **options: Any,
+    ) -> bool:
+        """Delete a container.
+
+        Args:
+            container_id: Container ID.
+            **options: Additional options.
+
+        Returns:
+            True if deleted, False if not found.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.delete(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return False
+            response.raise_for_status()
+
+        return True
+
+    # === Memory Storage ===
+
+    async def store(
+        self,
+        container_id: str,
+        request: StoreRequest,
+        **options: Any,
+    ) -> StoreResult:
+        """Store memory with optional inference.
+
+        Which strategies run is determined by:
+        1. How the container was configured (strategy -> namespace field mapping)
+        2. Which namespace fields are present in the request
+
+        Example:
+            Container has SEMANTIC scoped to "user_id", SUMMARY scoped to "session_id".
+            - namespace={"user_id": "123"} -> Runs SEMANTIC only
+            - namespace={"user_id": "123", "session_id": "abc"} -> Runs both
+
+        Args:
+            container_id: Target container.
+            request: Store request with namespace values.
+            **options: Additional options.
+
+        Returns:
+            Store result.
+        """
+        # Build request body
+        body: dict[str, Any] = {
+            "payload_type": request.payload_type.value,
+            "infer": request.infer,
+        }
+
+        if request.payload_type == PayloadType.CONVERSATIONAL:
+            if not request.messages:
+                raise MemoryError("Messages required for conversational payload")
+            body["messages"] = [m.to_dict() for m in request.messages]
+        else:
+            if not request.structured_data:
+                raise MemoryError("Structured data required for data payload")
+            body["structured_data"] = request.structured_data
+
+        # Namespace determines which strategies run based on container config
+        body["namespace"] = request.namespace.to_dict()
+
+        if request.metadata:
+            body["metadata"] = request.metadata
+        if request.tags:
+            body["tags"] = request.tags
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.inference_timeout
+            if request.infer
+            else self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        return StoreResult(
+            session_id=result.get("session_id"),
+            working_memory_id=result.get("working_memory_id"),
+        )
+
+    # === Working Memory ===
+
+    async def get_working_memory(
+        self,
+        container_id: str,
+        session_id: str | None = None,
+        namespace: Namespace | None = None,
+        limit: int = 50,
+        offset: int = 0,
+        **options: Any,
+    ) -> list[Message]:
+        """Get working memory messages.
+
+        Args:
+            container_id: Container ID.
+            session_id: Optional session filter.
+            namespace: Optional namespace filter.
+            limit: Maximum number of messages to return.
+            offset: Number of messages to skip.
+            **options: Additional options.
+
+        Returns:
+            List of messages.
+        """
+        search_body: dict[str, Any] = {
+            "query": {"match_all": {}},
+            "size": limit,
+            "from": offset,
+            "sort": [{"created_time": "asc"}],
+        }
+
+        filters = []
+        if session_id:
+            filters.append({"term": {"session_id": session_id}})
+        if namespace:
+            for key, value in namespace.values.items():
+                filters.append({"term": {f"namespace.{key}": value}})
+
+        if filters:
+            search_body["query"] = {"bool": {"filter": filters}}
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/working/_search",
+                json=search_body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        messages = []
+        for hit in result.get("hits", {}).get("hits", []):
+            source = hit["_source"]
+            content_parts = source.get("content", [])
+            content = content_parts[0].get("text", "") if content_parts else ""
+            messages.append(
+                Message(
+                    role=source.get("role", "user"),
+                    content=content,
+                    timestamp=datetime.fromtimestamp(source["created_time"] / 1000)
+                    if source.get("created_time")
+                    else None,
+                )
+            )
+
+        return messages
+
+    async def clear_working_memory(
+        self,
+        container_id: str,
+        session_id: str | None = None,
+        namespace: Namespace | None = None,
+        **options: Any,
+    ) -> int:
+        """Clear working memory.
+
+        Deletes working memory messages matching the filter criteria.
+        Uses delete-by-query internally.
+
+        Args:
+            container_id: Container ID.
+            session_id: Optional session filter.
+            namespace: Optional namespace filter.
+            **options: Backend-specific options.
+
+        Returns:
+            Number of messages deleted.
+        """
+        # Build query for delete-by-query
+        filters = []
+        if session_id:
+            filters.append({"term": {"session_id": session_id}})
+        if namespace:
+            for key, value in namespace.values.items():
+                filters.append({"term": {f"namespace.{key}": value}})
+
+        if filters:
+            query: dict[str, Any] = {"bool": {"filter": filters}}
+        else:
+            query = {"match_all": {}}
+
+        return await self.delete_by_query(
+            container_id=container_id,
+            memory_type=MemoryType.WORKING,
+            query=query,
+            **options,
+        )
+
+    async def delete_by_query(
+        self,
+        container_id: str,
+        memory_type: MemoryType,
+        query: dict[str, Any],
+        **options: Any,
+    ) -> int:
+        """Delete memories matching an OpenSearch Query DSL query.
+
+        Provides full flexibility for complex deletion criteria.
+
+        Example:
+            ```python
+            # Delete all memories for a user older than 30 days
+            await client.delete_by_query(
+                container_id=container_id,
+                memory_type=MemoryType.WORKING,
+                query={
+                    "bool": {
+                        "must": [
+                            {"term": {"namespace.user_id": "user123"}},
+                            {"range": {"created_time": {"lt": "now-30d"}}}
+                        ]
+                    }
+                }
+            )
+            ```
+
+        Args:
+            container_id: Container ID.
+            memory_type: Memory type to delete from.
+            query: OpenSearch Query DSL query.
+            **options: Backend-specific options.
+
+        Returns:
+            Number of documents deleted.
+        """
+        body = {"query": query}
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.inference_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/{memory_type.value}/_delete_by_query",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        return result.get("deleted", 0)
+
+    # === Memory Recall (Phase 5) ===
+
+    async def recall(
+        self,
+        container_id: str,
+        query: str,
+        namespace: Namespace | None = None,
+        strategies: list[MemoryStrategy] | None = None,
+        min_score: float | None = None,
+        limit: int = 10,
+        after: datetime | None = None,
+        before: datetime | None = None,
+        **options: Any,
+    ) -> RecallResult:
+        """Semantic search over long-term memory.
+
+        Uses the ML API endpoint (NOT direct index access).
+        CRITICAL: Memory content is in the 'memory' field, not 'content'.
+
+        Example:
+            ```python
+            result = await client.recall(
+                container_id="container-123",
+                query="user preferences",
+                namespace=Namespace({"user_id": "alice-123"}),
+                strategies=[MemoryStrategy.SEMANTIC],
+                limit=5,
+            )
+            for item in result.items:
+                print(f"{item.content} (score: {item.score})")
+            ```
+
+        Args:
+            container_id: Container ID.
+            query: Search query text.
+            namespace: Optional namespace filter.
+            strategies: Filter by specific strategies.
+            min_score: Minimum similarity score.
+            limit: Maximum results to return.
+            after: Filter by created after timestamp.
+            before: Filter by created before timestamp.
+            **options: Additional options.
+
+        Returns:
+            RecallResult with matching memory entries.
+        """
+        # Build neural query
+        search_body: dict[str, Any] = {
+            "query": {
+                "neural": {
+                    "memory_embedding": {
+                        "query_text": query,
+                        "model_id": self._config.embedding_model_id,
+                        "k": limit,
+                    },
+                },
+            },
+            "size": limit,
+            "_source": ["memory", "strategy_type", "namespace", "created_time"],
+        }
+
+        # Add filters
+        filters = []
+        if namespace:
+            for key, value in namespace.values.items():
+                filters.append({"term": {f"namespace.{key}": value}})
+        if strategies:
+            filters.append({"terms": {"strategy_type": [s.value for s in strategies]}})
+        if after:
+            filters.append(
+                {"range": {"created_time": {"gte": int(after.timestamp() * 1000)}}}
+            )
+        if before:
+            filters.append(
+                {"range": {"created_time": {"lte": int(before.timestamp() * 1000)}}}
+            )
+
+        if filters:
+            search_body["query"] = {
+                "bool": {
+                    "must": [search_body["query"]],
+                    "filter": filters,
+                },
+            }
+
+        if min_score:
+            search_body["min_score"] = min_score
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/long-term/_search",
+                json=search_body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        items = []
+        for hit in result.get("hits", {}).get("hits", []):
+            source = hit["_source"]
+            items.append(
+                MemoryEntry(
+                    id=hit["_id"],
+                    content=source.get("memory", ""),  # CRITICAL: field is "memory"
+                    strategy=MemoryStrategy(source["strategy_type"])
+                    if source.get("strategy_type")
+                    else None,
+                    score=hit.get("_score", 0.0),
+                    namespace=source.get("namespace", {}),
+                    created_at=datetime.fromtimestamp(source["created_time"] / 1000)
+                    if source.get("created_time")
+                    else None,
+                )
+            )
+
+        return RecallResult(
+            items=items,
+            total=result.get("hits", {}).get("total", {}).get("value", len(items)),
+            query=query,
+            took_ms=result.get("took", 0),
+        )
+
+    async def get_memory(
+        self,
+        container_id: str,
+        memory_id: str,
+        memory_type: MemoryType,
+        **options: Any,
+    ) -> MemoryEntry | None:
+        """Get a specific memory by ID.
+
+        Args:
+            container_id: Container ID.
+            memory_id: Memory document ID.
+            memory_type: Memory type (WORKING or LONG_TERM).
+            **options: Backend-specific options.
+
+        Returns:
+            Memory entry or None if not found.
+
+        Raises:
+            ValueError: If memory_type is SESSIONS or HISTORY.
+        """
+        if memory_type == MemoryType.SESSIONS:
+            raise ValueError("Use get_session() for session retrieval")
+        if memory_type == MemoryType.HISTORY:
+            raise ValueError("Use get_history_entry() for history retrieval")
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.get(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/{memory_type.value}/{memory_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return None
+            response.raise_for_status()
+            data = response.json()
+
+        return self._parse_memory_entry(memory_id, data, memory_type)
+
+    async def delete_memory(
+        self,
+        container_id: str,
+        memory_id: str,
+        memory_type: MemoryType,
+        **options: Any,
+    ) -> bool:
+        """Delete a specific memory by ID.
+
+        Args:
+            container_id: Container ID.
+            memory_id: Memory document ID.
+            memory_type: Memory type (WORKING, LONG_TERM).
+            **options: Backend-specific options.
+
+        Returns:
+            True if deleted, False if not found.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.delete(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/{memory_type.value}/{memory_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return False
+            response.raise_for_status()
+
+        return True
+
+    async def update_memory(
+        self,
+        container_id: str,
+        memory_id: str,
+        memory_type: MemoryType,
+        *,
+        memory: str | None = None,
+        tags: dict[str, str] | None = None,
+        **options: Any,
+    ) -> MemoryEntry:
+        """Update a specific memory.
+
+        Note: History memory type does NOT support updates.
+
+        Args:
+            container_id: Container ID.
+            memory_id: Memory document ID.
+            memory_type: Memory type (working, long-term, sessions).
+            memory: Updated memory content (for long-term).
+            tags: Updated tags.
+            **options: Additional options.
+
+        Returns:
+            Updated memory entry.
+
+        Raises:
+            MemoryError: If trying to update history.
+        """
+        if memory_type == MemoryType.HISTORY:
+            raise MemoryError("History memory does not support updates")
+
+        body: dict[str, Any] = {}
+        if memory is not None:
+            body["memory"] = memory
+        if tags is not None:
+            body["tags"] = tags
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.put(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/{memory_type.value}/{memory_id}",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            data = response.json()
+
+        return MemoryEntry(
+            id=memory_id,
+            content=data.get("memory", ""),
+            strategy=MemoryStrategy(data["strategy_type"])
+            if data.get("strategy_type")
+            else None,
+            namespace=data.get("namespace", {}),
+            metadata=data.get("tags", {}),
+        )
+
+    async def delete_memories(
+        self,
+        container_id: str,
+        session_id: str | None = None,
+        namespace: Namespace | None = None,
+        before: datetime | None = None,
+        **options: Any,
+    ) -> int:
+        """Delete memories by filter.
+
+        This is a convenience wrapper around delete_by_query.
+
+        Args:
+            container_id: Container ID.
+            session_id: Filter by session.
+            namespace: Filter by namespace.
+            before: Delete memories created before this timestamp.
+            **options: Additional options (memory_type defaults to WORKING).
+
+        Returns:
+            Number of memories deleted.
+        """
+        # Build query from filters
+        filters = []
+
+        if session_id:
+            filters.append({"term": {"session_id": session_id}})
+        if namespace:
+            for key, value in namespace.values.items():
+                filters.append({"term": {f"namespace.{key}": value}})
+        if before:
+            filters.append(
+                {"range": {"created_time": {"lt": int(before.timestamp() * 1000)}}}
+            )
+
+        if filters:
+            query: dict[str, Any] = {"bool": {"filter": filters}}
+        else:
+            query = {"match_all": {}}
+
+        # Default to working memory if not specified
+        memory_type = options.pop("memory_type", MemoryType.WORKING)
+
+        return await self.delete_by_query(
+            container_id=container_id,
+            memory_type=memory_type,
+            query=query,
+            **options,
+        )
+
+    # === Session Management (Phase 6) ===
+
+    async def create_session(
+        self,
+        container_id: str,
+        *,
+        session_id: str | None = None,
+        summary: str | None = None,
+        namespace: Namespace | None = None,
+        metadata: dict[str, Any] | None = None,
+        **options: Any,
+    ) -> SessionInfo:
+        """Create a new session.
+
+        Args:
+            container_id: Container ID.
+            session_id: Custom session ID (auto-generated if not provided).
+            summary: Session summary text.
+            namespace: Session namespace.
+            metadata: Custom metadata (stored as additional_info).
+            **options: Additional options.
+
+        Returns:
+            Created session info.
+        """
+        body: dict[str, Any] = {}
+        if session_id:
+            body["session_id"] = session_id
+        if summary:
+            body["summary"] = summary
+        if namespace:
+            body["namespace"] = namespace.values
+        if metadata:
+            body["additional_info"] = metadata
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/sessions",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        return SessionInfo(
+            id=result.get("session_id"),
+            container_id=container_id,
+            summary=summary,
+            namespace=namespace.values if namespace else {},
+            metadata=metadata or {},
+        )
+
+    async def get_session(
+        self,
+        container_id: str,
+        session_id: str,
+        include_messages: bool = False,
+        message_limit: int = 50,
+        **options: Any,
+    ) -> SessionInfo | None:
+        """Get session by ID.
+
+        Args:
+            container_id: Container ID.
+            session_id: Session ID.
+            include_messages: Whether to include session messages.
+            message_limit: Max messages to include.
+            **options: Additional options.
+
+        Returns:
+            Session info or None if not found.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.get(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/sessions/{session_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return None
+            response.raise_for_status()
+            data = response.json()
+
+        session = SessionInfo(
+            id=session_id,
+            container_id=container_id,
+            summary=data.get("summary"),
+            namespace=data.get("namespace", {}),
+            metadata=data.get("additional_info", {}),
+            started_at=datetime.fromtimestamp(data["created_time"] / 1000)
+            if data.get("created_time")
+            else None,
+        )
+
+        if include_messages:
+            session.messages = await self.get_working_memory(
+                container_id=container_id,
+                session_id=session_id,
+                limit=message_limit,
+            )
+
+        return session
+
+    async def list_sessions(
+        self,
+        container_id: str,
+        namespace: Namespace | None = None,
+        limit: int = 100,
+        **options: Any,
+    ) -> list[SessionInfo]:
+        """List sessions.
+
+        Args:
+            container_id: Container ID.
+            namespace: Optional namespace filter.
+            limit: Maximum sessions to return.
+            **options: Additional options.
+
+        Returns:
+            List of session info.
+        """
+        search_body: dict[str, Any] = {
+            "query": {"match_all": {}},
+            "size": limit,
+        }
+
+        if namespace:
+            filters = [
+                {"term": {f"namespace.{k}": v}} for k, v in namespace.values.items()
+            ]
+            search_body["query"] = {"bool": {"filter": filters}}
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/sessions/_search",
+                json=search_body,
+                auth=self._auth,
+            )
+            # Handle 500 error when sessions index doesn't exist yet
+            # OpenSearch returns "index must not be null" when no sessions have been created
+            if response.status_code == 500:
+                try:
+                    error_body = response.json()
+                    error_reason = error_body.get("error", {}).get("reason", "")
+                    if "index must not be null" in error_reason:
+                        return []
+                except Exception:
+                    pass
+            response.raise_for_status()
+            result = response.json()
+
+        sessions = []
+        for hit in result.get("hits", {}).get("hits", []):
+            source = hit["_source"]
+            sessions.append(
+                SessionInfo(
+                    id=hit["_id"],
+                    container_id=container_id,
+                    summary=source.get("summary"),
+                    namespace=source.get("namespace", {}),
+                    metadata=source.get("additional_info", {}),
+                    started_at=datetime.fromtimestamp(source["created_time"] / 1000)
+                    if source.get("created_time")
+                    else None,
+                )
+            )
+
+        return sessions
+
+    async def update_session(
+        self,
+        container_id: str,
+        session_id: str,
+        *,
+        summary: str | None = None,
+        metadata: dict[str, Any] | None = None,
+        **options: Any,
+    ) -> SessionInfo:
+        """Update a session.
+
+        Args:
+            container_id: Container ID.
+            session_id: Session ID.
+            summary: Updated summary text.
+            metadata: Updated metadata (additional_info).
+            **options: Additional options.
+
+        Returns:
+            Updated session info.
+        """
+        body: dict[str, Any] = {}
+        if summary is not None:
+            body["summary"] = summary
+        if metadata is not None:
+            body["additional_info"] = metadata
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.put(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/sessions/{session_id}",
+                json=body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            data = response.json()
+
+        return SessionInfo(
+            id=session_id,
+            container_id=container_id,
+            summary=data.get("summary"),
+            namespace=data.get("namespace", {}),
+            metadata=data.get("additional_info", {}),
+        )
+
+    async def delete_session(
+        self,
+        container_id: str,
+        session_id: str,
+        **options: Any,
+    ) -> bool:
+        """Delete a session.
+
+        Args:
+            container_id: Container ID.
+            session_id: Session ID.
+            **options: Additional options.
+
+        Returns:
+            True if deleted, False if not found.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.delete(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/sessions/{session_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return False
+            response.raise_for_status()
+
+        return True
+
+    # === History (Audit Trail - Phase 6) ===
+
+    async def get_history_entry(
+        self,
+        container_id: str,
+        history_id: str,
+        **options: Any,
+    ) -> HistoryEntry | None:
+        """Get a specific history entry by ID.
+
+        History entries are READ-ONLY audit trail records.
+
+        Args:
+            container_id: Container ID.
+            history_id: History entry ID.
+            **options: Additional options.
+
+        Returns:
+            History entry or None if not found.
+        """
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.get(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/history/{history_id}",
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return None
+            response.raise_for_status()
+            data = response.json()
+
+        return self._parse_history_entry(history_id, container_id, data)
+
+    async def list_history(
+        self,
+        container_id: str,
+        memory_id: str | None = None,
+        namespace: Namespace | None = None,
+        limit: int = 100,
+        **options: Any,
+    ) -> list[HistoryEntry]:
+        """List history entries.
+
+        Args:
+            container_id: Container ID.
+            memory_id: Filter by specific memory ID.
+            namespace: Filter by namespace.
+            limit: Maximum entries to return.
+            **options: Additional options.
+
+        Returns:
+            List of history entries (most recent first).
+        """
+        search_body: dict[str, Any] = {
+            "query": {"match_all": {}},
+            "size": limit,
+            "sort": [{"created_time": "desc"}],
+        }
+
+        filters = []
+        if memory_id:
+            filters.append({"term": {"memory_id": memory_id}})
+        if namespace:
+            for key, value in namespace.values.items():
+                filters.append({"term": {f"namespace.{key}": value}})
+
+        if filters:
+            search_body["query"] = {"bool": {"filter": filters}}
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/history/_search",
+                json=search_body,
+                auth=self._auth,
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        entries = []
+        for hit in result.get("hits", {}).get("hits", []):
+            entry = self._parse_history_entry(hit["_id"], container_id, hit["_source"])
+            entries.append(entry)
+
+        return entries
+
+    # === Statistics (Phase 6) ===
+
+    async def get_stats(
+        self,
+        container_id: str,
+        **options: Any,
+    ) -> MemoryStats:
+        """Get container statistics.
+
+        Args:
+            container_id: Container ID.
+            **options: Additional options.
+
+        Returns:
+            Memory statistics for the container.
+
+        Raises:
+            ContainerNotFoundError: If container doesn't exist.
+        """
+        container = await self.get_container(container_id)
+        if not container:
+            raise ContainerNotFoundError(container_id=container_id)
+
+        # Count working memory
+        working_count = await self._count_memories(container_id, MemoryType.WORKING)
+
+        # Count long-term memory
+        long_term_count = await self._count_memories(container_id, MemoryType.LONG_TERM)
+
+        # Count sessions
+        session_count = await self._count_memories(container_id, MemoryType.SESSIONS)
+
+        # Get strategy breakdown
+        breakdown: dict[MemoryStrategy, int] = {}
+        for strategy in MemoryStrategy:
+            count = await self._count_memories(
+                container_id,
+                MemoryType.LONG_TERM,
+                strategy=strategy,
+            )
+            if count > 0:
+                breakdown[strategy] = count
+
+        return MemoryStats(
+            container_id=container_id,
+            container_name=container.name,
+            working_memory_count=working_count,
+            long_term_memory_count=long_term_count,
+            session_count=session_count,
+            strategies_breakdown=breakdown,
+        )
+
+    async def _count_memories(
+        self,
+        container_id: str,
+        memory_type: MemoryType,
+        strategy: MemoryStrategy | None = None,
+    ) -> int:
+        """Count memories by type and optional strategy.
+
+        Args:
+            container_id: Container ID.
+            memory_type: Memory type to count.
+            strategy: Optional strategy filter.
+
+        Returns:
+            Number of memories matching criteria.
+        """
+        search_body: dict[str, Any] = {
+            "query": {"match_all": {}},
+            "size": 0,
+        }
+
+        if strategy:
+            search_body["query"] = {"term": {"strategy_type": strategy.value}}
+
+        async with httpx.AsyncClient(
+            verify=self._config.verify_certs,
+            timeout=self._config.connect_timeout,
+        ) as client:
+            response = await client.post(
+                f"{self._base_url}/_plugins/_ml/memory_containers/{container_id}/memories/{memory_type.value}/_search",
+                json=search_body,
+                auth=self._auth,
+            )
+            if response.status_code == 404:
+                return 0
+            # Handle 500 error when sessions index doesn't exist yet
+            # OpenSearch returns "index must not be null" when no sessions have been created
+            if response.status_code == 500:
+                try:
+                    error_body = response.json()
+                    error_reason = error_body.get("error", {}).get("reason", "")
+                    if "index must not be null" in error_reason:
+                        return 0
+                except Exception:
+                    pass
+            response.raise_for_status()
+            result = response.json()
+
+        return result.get("hits", {}).get("total", {}).get("value", 0)
+
+    # === Helpers ===
+
+    def _parse_container_info(
+        self, container_id: str, data: dict[str, Any]
+    ) -> ContainerInfo:
+        """Parse container info from API response.
+
+        Args:
+            container_id: Container ID.
+            data: Raw API response data.
+
+        Returns:
+            Parsed ContainerInfo object.
+        """
+        config = data.get("configuration", {})
+        strategies = [
+            MemoryStrategy(s["type"]) for s in config.get("strategies", [])
+        ]
+
+        return ContainerInfo(
+            id=container_id,
+            name=data.get("name", ""),
+            description=data.get("description"),
+            strategies=strategies,
+            embedding_model_id=config.get("embedding_model_id"),
+            llm_model_id=config.get("llm_id"),
+            created_at=datetime.fromtimestamp(data["created_time"] / 1000)
+            if data.get("created_time")
+            else None,
+            updated_at=datetime.fromtimestamp(data["last_updated_time"] / 1000)
+            if data.get("last_updated_time")
+            else None,
+        )
+
+    def _parse_memory_entry(
+        self,
+        memory_id: str,
+        data: dict[str, Any],
+        memory_type: MemoryType,
+    ) -> MemoryEntry:
+        """Parse memory entry from API response.
+
+        Handles both working memory and long-term memory formats.
+        CRITICAL: Long-term memory content is in 'memory' field, not 'content'.
+
+        Args:
+            memory_id: Memory document ID.
+            data: Raw API response data.
+            memory_type: Type of memory being parsed.
+
+        Returns:
+            Parsed MemoryEntry object.
+        """
+        # Long-term memory has 'memory' field with extracted content
+        # Working memory has 'messages' array with conversation
+        if memory_type == MemoryType.LONG_TERM:
+            content = data.get("memory", "")
+            strategy = (
+                MemoryStrategy(data["strategy_type"])
+                if data.get("strategy_type")
+                else None
+            )
+        else:
+            # Working memory: extract text from messages
+            messages = data.get("messages", [])
+            content_parts = []
+            for msg in messages:
+                msg_content = msg.get("content", [])
+                for part in msg_content:
+                    if part.get("type") == "text":
+                        content_parts.append(
+                            f"[{msg.get('role', 'unknown')}]: {part.get('text', '')}"
+                        )
+            content = "\n".join(content_parts)
+            strategy = None
+
+        return MemoryEntry(
+            id=memory_id,
+            content=content,
+            strategy=strategy,
+            score=0.0,
+            namespace=data.get("namespace", {}),
+            created_at=datetime.fromtimestamp(data["created_time"] / 1000)
+            if data.get("created_time")
+            else None,
+            metadata=data.get("tags", {}),
+        )
+
+    def _parse_history_entry(
+        self,
+        history_id: str,
+        container_id: str,
+        data: dict[str, Any],
+    ) -> HistoryEntry:
+        """Parse history entry from API response.
+
+        Args:
+            history_id: History entry ID.
+            container_id: Parent container ID.
+            data: Raw API response data.
+
+        Returns:
+            Parsed HistoryEntry object.
+        """
+        return HistoryEntry(
+            id=history_id,
+            memory_id=data.get("memory_id", ""),
+            container_id=container_id,
+            action=HistoryAction(data["action"])
+            if data.get("action")
+            else HistoryAction.ADD,
+            owner_id=data.get("owner_id"),
+            before=data.get("before"),
+            after=data.get("after"),
+            namespace=data.get("namespace", {}),
+            tags=data.get("tags", {}),
+            created_at=datetime.fromtimestamp(data["created_time"] / 1000)
+            if data.get("created_time")
+            else None,
+        )