agno 2.2.13__py3-none-any.whl → 2.4.3__py3-none-any.whl

agno/learn/stores/user_memory.py

@@ -0,0 +1,1495 @@
+"""
+User Memory Store
+=================
+Storage backend for User Memory learning type.
+
+Stores unstructured observations about users that don't fit into
+structured profile fields. These are long-term memories that persist
+across sessions.
+
+Key Features:
+- Background extraction from conversations
+- Agent tools for in-conversation updates
+- Multi-user isolation (each user has their own memories)
+- Add, update, delete memory operations
+
+Scope:
+- Memories are retrieved by user_id only
+- agent_id/team_id stored in DB columns for audit trail
+- agent_id/team_id stored on individual memories for granular audit
+
+Supported Modes:
+- ALWAYS: Automatic extraction after conversations
+- AGENTIC: Agent calls update_user_memory tool directly
+"""
+
+import uuid
+from copy import deepcopy
+from dataclasses import dataclass, field
+from os import getenv
+from textwrap import dedent
+from typing import Any, Callable, List, Optional, Union
+
+from agno.learn.config import LearningMode, UserMemoryConfig
+from agno.learn.schemas import Memories
+from agno.learn.stores.protocol import LearningStore
+from agno.learn.utils import from_dict_safe, to_dict_safe
+from agno.utils.log import (
+    log_debug,
+    log_warning,
+    set_log_level_to_debug,
+    set_log_level_to_info,
+)
+
+try:
+    from agno.db.base import AsyncBaseDb, BaseDb
+    from agno.models.message import Message
+    from agno.tools.function import Function
+except ImportError:
+    pass
+
+
+@dataclass
+class UserMemoryStore(LearningStore):
+    """Storage backend for User Memory learning type.
+
+    Memories are retrieved by user_id only - all agents sharing the same DB
+    will see the same memories for a given user. agent_id and team_id are
+    stored for audit purposes (both at DB column level and on individual memories).
+
+    Args:
+        config: UserMemoryConfig with all settings including db and model.
+        debug_mode: Enable debug logging.
+    """
+
+    config: UserMemoryConfig = field(default_factory=UserMemoryConfig)
+    debug_mode: bool = False
+
+    # State tracking (internal)
+    memories_updated: bool = field(default=False, init=False)
+    _schema: Any = field(default=None, init=False)
+
+    def __post_init__(self):
+        self._schema = self.config.schema or Memories
+
+        if self.config.mode == LearningMode.PROPOSE:
+            log_warning("UserMemoryStore does not support PROPOSE mode.")
+        elif self.config.mode == LearningMode.HITL:
+            log_warning("UserMemoryStore does not support HITL mode.")
+
+    # =========================================================================
+    # LearningStore Protocol Implementation
+    # =========================================================================
+
+    @property
+    def learning_type(self) -> str:
+        """Unique identifier for this learning type."""
+        return "user_memory"
+
+    @property
+    def schema(self) -> Any:
+        """Schema class used for memories."""
+        return self._schema
+
+    def recall(self, user_id: str, **kwargs) -> Optional[Any]:
+        """Retrieve memories from storage.
+
+        Args:
+            user_id: The user to retrieve memories for (required).
+            **kwargs: Additional context (ignored).
+
+        Returns:
+            Memories, or None if not found.
+        """
+        if not user_id:
+            return None
+        return self.get(user_id=user_id)
+
+    async def arecall(self, user_id: str, **kwargs) -> Optional[Any]:
+        """Async version of recall."""
+        if not user_id:
+            return None
+        return await self.aget(user_id=user_id)
+
+    def process(
+        self,
+        messages: List[Any],
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+        **kwargs,
+    ) -> None:
+        """Extract memories from messages.
+
+        Args:
+            messages: Conversation messages to analyze.
+            user_id: The user to update memories for (required).
+            agent_id: Agent context (stored for audit).
+            team_id: Team context (stored for audit).
+            **kwargs: Additional context (ignored).
+        """
+        # process only supported in ALWAYS mode
+        # for programmatic extraction, use extract_and_save directly
+        if self.config.mode != LearningMode.ALWAYS:
+            return
+
+        if not user_id or not messages:
+            return
+
+        self.extract_and_save(
+            messages=messages,
+            user_id=user_id,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+    async def aprocess(
+        self,
+        messages: List[Any],
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+        **kwargs,
+    ) -> None:
+        """Async version of process."""
+        if self.config.mode != LearningMode.ALWAYS:
+            return
+
+        if not user_id or not messages:
+            return
+
+        await self.aextract_and_save(
+            messages=messages,
+            user_id=user_id,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+    def build_context(self, data: Any) -> str:
+        """Build context for the agent.
+
+        Formats memories data for injection into the agent's system prompt.
+        Designed to enable natural, personalized responses without meta-commentary
+        about memory systems.
+
+        Args:
+            data: Memories data from recall().
+
+        Returns:
+            Context string to inject into the agent's system prompt.
+        """
+        # Build tool documentation based on what's enabled
+        tool_docs = self._build_tool_documentation()
+
+        if not data:
+            if self._should_expose_tools:
+                return (
+                    dedent("""\
+                    <user_memory>
+                    No memories saved about this user yet.
+
+                    """)
+                    + tool_docs
+                    + dedent("""
+                    </user_memory>""")
+                )
+            return ""
+
+        # Build memories section
+        memories_text = None
+        if hasattr(data, "get_memories_text"):
+            memories_text = data.get_memories_text()
+        elif hasattr(data, "memories") and data.memories:
+            memories_text = "\n".join(f"- {m.get('content', str(m))}" for m in data.memories)
+
+        if not memories_text:
+            if self._should_expose_tools:
+                return (
+                    dedent("""\
+                    <user_memory>
+                    No memories saved about this user yet.
+
+                    """)
+                    + tool_docs
+                    + dedent("""
+                    </user_memory>""")
+                )
+            return ""
+
+        context = "<user_memory>\n"
+        context += memories_text + "\n"
+
+        context += dedent("""
+            <memory_application_guidelines>
+            Apply this knowledge naturally - respond as if you inherently know this information,
+            exactly as a colleague would recall shared history without narrating their thought process.
+
+            - Selectively apply memories based on relevance to the current query
+            - Never say "based on my memory" or "I remember that" - just use the information naturally
+            - Current conversation always takes precedence over stored memories
+            - Use memories to calibrate tone, depth, and examples without announcing it
+            </memory_application_guidelines>""")
+
+        if self._should_expose_tools:
+            context += (
+                dedent("""
+
+            <memory_updates>
+
+            """)
+                + tool_docs
+                + dedent("""
+            </memory_updates>""")
+            )
+
+        context += "\n</user_memory>"
+
+        return context
+
+    def _build_tool_documentation(self) -> str:
+        """Build documentation for available memory tools.
+
+        Returns:
+            String documenting which tools are available and when to use them.
+        """
+        docs = []
+
+        if self.config.agent_can_update_memories:
+            docs.append(
+                "Use `update_user_memory` to save observations, preferences, and context about this user "
+                "that would help personalize future conversations or avoid asking the same questions."
+            )
+
+        return "\n\n".join(docs) if docs else ""
+
+    def get_tools(
+        self,
+        user_id: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+        **kwargs,
+    ) -> List[Callable]:
+        """Get tools to expose to agent.
+
+        Args:
+            user_id: The user context (required for tool to work).
+            agent_id: Agent context (stored for audit).
+            team_id: Team context (stored for audit).
+            **kwargs: Additional context (ignored).
+
+        Returns:
+            List containing update_user_memory tool if enabled.
+        """
+        if not user_id or not self._should_expose_tools:
+            return []
+        return self.get_agent_tools(
+            user_id=user_id,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+    async def aget_tools(
+        self,
+        user_id: Optional[str] = None,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+        **kwargs,
+    ) -> List[Callable]:
+        """Async version of get_tools."""
+        if not user_id or not self._should_expose_tools:
+            return []
+        return await self.aget_agent_tools(
+            user_id=user_id,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+    @property
+    def was_updated(self) -> bool:
+        """Check if memories were updated in last operation."""
+        return self.memories_updated
+
+    @property
+    def _should_expose_tools(self) -> bool:
+        """Check if tools should be exposed to the agent.
+
+        Returns True if either:
+        - mode is AGENTIC (tools are the primary way to update memory), OR
+        - enable_agent_tools is explicitly True
+        """
+        return self.config.mode == LearningMode.AGENTIC or self.config.enable_agent_tools
+
+    # =========================================================================
+    # Properties
+    # =========================================================================
+
+    @property
+    def db(self) -> Optional[Union["BaseDb", "AsyncBaseDb"]]:
+        """Database backend."""
+        return self.config.db
+
+    @property
+    def model(self):
+        """Model for extraction."""
+        return self.config.model
+
+    # =========================================================================
+    # Debug/Logging
+    # =========================================================================
+
+    def set_log_level(self):
+        """Set log level based on debug_mode or environment variable."""
+        if self.debug_mode or getenv("AGNO_DEBUG", "false").lower() == "true":
+            self.debug_mode = True
+            set_log_level_to_debug()
+        else:
+            set_log_level_to_info()
+
+    # =========================================================================
+    # Agent Tools
+    # =========================================================================
+
+    def get_agent_tools(
+        self,
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> List[Callable]:
+        """Get the tools to expose to the agent.
+
+        Args:
+            user_id: The user to update (required).
+            agent_id: Agent context (stored for audit).
+            team_id: Team context (stored for audit).
+
+        Returns:
+            List of callable tools based on config settings.
+        """
+        tools = []
+
+        # Memory update tool (delegates to extraction)
+        if self.config.agent_can_update_memories:
+
+            def update_user_memory(task: str) -> str:
+                """Save or update information about this user for future conversations.
+
+                Use this when you learn something worth remembering - information that would
+                help personalize future interactions or provide continuity across sessions.
+
+                Args:
+                    task: What to save, update, or remove. Be specific and factual.
+                          Good examples:
+                          - "User is a senior engineer at Stripe working on payments"
+                          - "Prefers concise responses without lengthy explanations"
+                          - "Update: User moved from NYC to London"
+                          - "Remove the memory about their old job at Acme"
+                          Bad examples:
+                          - "User seems nice" (too vague)
+                          - "Had a meeting today" (not durable)
+
+                Returns:
+                    Confirmation of what was saved/updated.
+                """
+                return self.run_memories_update(
+                    task=task,
+                    user_id=user_id,
+                    agent_id=agent_id,
+                    team_id=team_id,
+                )
+
+            tools.append(update_user_memory)
+
+        return tools
+
+    async def aget_agent_tools(
+        self,
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> List[Callable]:
+        """Get the async tools to expose to the agent."""
+        tools = []
+
+        if self.config.agent_can_update_memories:
+
+            async def update_user_memory(task: str) -> str:
+                """Save or update information about this user for future conversations.
+
+                Use this when you learn something worth remembering - information that would
+                help personalize future interactions or provide continuity across sessions.
+
+                Args:
+                    task: What to save, update, or remove. Be specific and factual.
+                          Good examples:
+                          - "User is a senior engineer at Stripe working on payments"
+                          - "Prefers concise responses without lengthy explanations"
+                          - "Update: User moved from NYC to London"
+                          - "Remove the memory about their old job at Acme"
+                          Bad examples:
+                          - "User seems nice" (too vague)
+                          - "Had a meeting today" (not durable)
+
+                Returns:
+                    Confirmation of what was saved/updated.
+                """
+                return await self.arun_memories_update(
+                    task=task,
+                    user_id=user_id,
+                    agent_id=agent_id,
+                    team_id=team_id,
+                )
+
+            tools.append(update_user_memory)
+
+        return tools
+
+    # =========================================================================
+    # Read Operations
+    # =========================================================================
+
+    def get(self, user_id: str) -> Optional[Any]:
+        """Retrieve memories by user_id.
+
+        Args:
+            user_id: The unique user identifier.
+
+        Returns:
+            Memories as schema instance, or None if not found.
+        """
+        if not self.db:
+            return None
+
+        try:
+            result = self.db.get_learning(
+                learning_type=self.learning_type,
+                user_id=user_id,
+            )
+
+            if result and result.get("content"):  # type: ignore[union-attr]
+                return from_dict_safe(self.schema, result["content"])  # type: ignore[index]
+
+            return None
+
+        except Exception as e:
+            log_debug(f"UserMemoryStore.get failed for user_id={user_id}: {e}")
+            return None
+
+    async def aget(self, user_id: str) -> Optional[Any]:
+        """Async version of get."""
+        if not self.db:
+            return None
+
+        try:
+            if isinstance(self.db, AsyncBaseDb):
+                result = await self.db.get_learning(
+                    learning_type=self.learning_type,
+                    user_id=user_id,
+                )
+            else:
+                result = self.db.get_learning(
+                    learning_type=self.learning_type,
+                    user_id=user_id,
+                )
+
+            if result and result.get("content"):
+                return from_dict_safe(self.schema, result["content"])
+
+            return None
+
+        except Exception as e:
+            log_debug(f"UserMemoryStore.aget failed for user_id={user_id}: {e}")
+            return None
+
+    # =========================================================================
+    # Write Operations
+    # =========================================================================
+
+    def save(
+        self,
+        user_id: str,
+        memories: Any,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> None:
+        """Save or update memories.
+
+        Args:
+            user_id: The unique user identifier.
+            memories: The memories data to save.
+            agent_id: Agent context (stored in DB column for audit).
+            team_id: Team context (stored in DB column for audit).
+        """
+        if not self.db or not memories:
+            return
+
+        try:
+            content = to_dict_safe(memories)
+            if not content:
+                return
+
+            self.db.upsert_learning(
+                id=self._build_memories_id(user_id=user_id),
+                learning_type=self.learning_type,
+                user_id=user_id,
+                agent_id=agent_id,
+                team_id=team_id,
+                content=content,
+            )
+            log_debug(f"UserMemoryStore.save: saved memories for user_id={user_id}")
+
+        except Exception as e:
+            log_debug(f"UserMemoryStore.save failed for user_id={user_id}: {e}")
+
+    async def asave(
+        self,
+        user_id: str,
+        memories: Any,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> None:
+        """Async version of save."""
+        if not self.db or not memories:
+            return
+
+        try:
+            content = to_dict_safe(memories)
+            if not content:
+                return
+
+            if isinstance(self.db, AsyncBaseDb):
+                await self.db.upsert_learning(
+                    id=self._build_memories_id(user_id=user_id),
+                    learning_type=self.learning_type,
+                    user_id=user_id,
+                    agent_id=agent_id,
+                    team_id=team_id,
+                    content=content,
+                )
+            else:
+                self.db.upsert_learning(
+                    id=self._build_memories_id(user_id=user_id),
+                    learning_type=self.learning_type,
+                    user_id=user_id,
+                    agent_id=agent_id,
+                    team_id=team_id,
+                    content=content,
+                )
+            log_debug(f"UserMemoryStore.asave: saved memories for user_id={user_id}")
+
+        except Exception as e:
+            log_debug(f"UserMemoryStore.asave failed for user_id={user_id}: {e}")
+
+    # =========================================================================
+    # Delete Operations
+    # =========================================================================
+
+    def delete(self, user_id: str) -> bool:
+        """Delete memories for a user.
+
+        Args:
+            user_id: The unique user identifier.
+
+        Returns:
+            True if deleted, False otherwise.
+        """
+        if not self.db:
+            return False
+
+        try:
+            memories_id = self._build_memories_id(user_id=user_id)
+            return self.db.delete_learning(id=memories_id)  # type: ignore[return-value]
+        except Exception as e:
+            log_debug(f"UserMemoryStore.delete failed for user_id={user_id}: {e}")
+            return False
+
+    async def adelete(self, user_id: str) -> bool:
+        """Async version of delete."""
+        if not self.db:
+            return False
+
+        try:
+            memories_id = self._build_memories_id(user_id=user_id)
+            if isinstance(self.db, AsyncBaseDb):
+                return await self.db.delete_learning(id=memories_id)
+            else:
+                return self.db.delete_learning(id=memories_id)
+        except Exception as e:
+            log_debug(f"UserMemoryStore.adelete failed for user_id={user_id}: {e}")
+            return False
+
+    def clear(
+        self,
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> None:
+        """Clear all memories for a user (reset to empty).
+
+        Args:
+            user_id: The unique user identifier.
+            agent_id: Agent context (stored for audit).
+            team_id: Team context (stored for audit).
+        """
+        if not self.db:
+            return
+
+        try:
+            empty_memories = self.schema(user_id=user_id)
+            self.save(user_id=user_id, memories=empty_memories, agent_id=agent_id, team_id=team_id)
+            log_debug(f"UserMemoryStore.clear: cleared memories for user_id={user_id}")
+        except Exception as e:
+            log_debug(f"UserMemoryStore.clear failed for user_id={user_id}: {e}")
+
+    async def aclear(
+        self,
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> None:
+        """Async version of clear."""
+        if not self.db:
+            return
+
+        try:
+            empty_memories = self.schema(user_id=user_id)
+            await self.asave(user_id=user_id, memories=empty_memories, agent_id=agent_id, team_id=team_id)
+            log_debug(f"UserMemoryStore.aclear: cleared memories for user_id={user_id}")
+        except Exception as e:
+            log_debug(f"UserMemoryStore.aclear failed for user_id={user_id}: {e}")
+
+    # =========================================================================
+    # Memory Operations
+    # =========================================================================
+
+    def add_memory(
+        self,
+        user_id: str,
+        memory: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+        **kwargs,
+    ) -> Optional[str]:
+        """Add a single memory.
+
+        Args:
+            user_id: The unique user identifier.
+            memory: The memory text to add.
+            agent_id: Agent that added this (stored for audit).
+            team_id: Team context (stored for audit).
+            **kwargs: Additional fields for the memory.
+
+        Returns:
+            The memory ID if added, None otherwise.
+        """
+        memories_data = self.get(user_id=user_id)
+
+        if memories_data is None:
+            memories_data = self.schema(user_id=user_id)
+
+        memory_id = None
+        if hasattr(memories_data, "add_memory"):
+            memory_id = memories_data.add_memory(memory, **kwargs)
+        elif hasattr(memories_data, "memories"):
+            memory_id = str(uuid.uuid4())[:8]
+            memory_entry = {"id": memory_id, "content": memory, **kwargs}
+            if agent_id:
+                memory_entry["added_by_agent"] = agent_id
+            if team_id:
+                memory_entry["added_by_team"] = team_id
+            memories_data.memories.append(memory_entry)
+
+        self.save(user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id)
+        log_debug(f"UserMemoryStore.add_memory: added memory for user_id={user_id}")
+
+        return memory_id
+
+    async def aadd_memory(
+        self,
+        user_id: str,
+        memory: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+        **kwargs,
+    ) -> Optional[str]:
+        """Async version of add_memory."""
+        memories_data = await self.aget(user_id=user_id)
+
+        if memories_data is None:
+            memories_data = self.schema(user_id=user_id)
+
+        memory_id = None
+        if hasattr(memories_data, "add_memory"):
+            memory_id = memories_data.add_memory(memory, **kwargs)
+        elif hasattr(memories_data, "memories"):
+            memory_id = str(uuid.uuid4())[:8]
+            memory_entry = {"id": memory_id, "content": memory, **kwargs}
+            if agent_id:
+                memory_entry["added_by_agent"] = agent_id
+            if team_id:
+                memory_entry["added_by_team"] = team_id
+            memories_data.memories.append(memory_entry)
+
+        await self.asave(user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id)
+        log_debug(f"UserMemoryStore.aadd_memory: added memory for user_id={user_id}")
+
+        return memory_id
+
+    # =========================================================================
+    # Extraction Operations
+    # =========================================================================
+
+    def extract_and_save(
+        self,
+        messages: List["Message"],
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> str:
+        """Extract memories from messages and save.
+
+        Args:
+            messages: Conversation messages to analyze.
+            user_id: The unique user identifier.
+            agent_id: Agent context (stored for audit).
+            team_id: Team context (stored for audit).
+
+        Returns:
+            Response from model.
+        """
+        if self.model is None:
+            log_warning("UserMemoryStore.extract_and_save: no model provided")
+            return "No model provided for memories extraction"
+
+        if not self.db:
+            log_warning("UserMemoryStore.extract_and_save: no database provided")
+            return "No DB provided for memories store"
+
+        log_debug("UserMemoryStore: Extracting memories", center=True)
+
+        self.memories_updated = False
+
+        existing_memories = self.get(user_id=user_id)
+        existing_data = self._memories_to_list(memories=existing_memories)
+
+        input_string = self._messages_to_input_string(messages=messages)
+
+        tools = self._get_extraction_tools(
+            user_id=user_id,
+            input_string=input_string,
+            existing_memories=existing_memories,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+        functions = self._build_functions_for_model(tools=tools)
+
+        messages_for_model = [
+            self._get_system_message(existing_data=existing_data),
+            *messages,
+        ]
+
+        model_copy = deepcopy(self.model)
+        response = model_copy.response(
+            messages=messages_for_model,
+            tools=functions,
+        )
+
+        if response.tool_executions:
+            self.memories_updated = True
+
+        log_debug("UserMemoryStore: Extraction complete", center=True)
+
+        return response.content or ("Memories updated" if self.memories_updated else "No updates needed")
+
+    async def aextract_and_save(
+        self,
+        messages: List["Message"],
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> str:
+        """Async version of extract_and_save."""
+        if self.model is None:
+            log_warning("UserMemoryStore.aextract_and_save: no model provided")
+            return "No model provided for memories extraction"
+
+        if not self.db:
+            log_warning("UserMemoryStore.aextract_and_save: no database provided")
+            return "No DB provided for memories store"
+
+        log_debug("UserMemoryStore: Extracting memories (async)", center=True)
+
+        self.memories_updated = False
+
+        existing_memories = await self.aget(user_id=user_id)
+        existing_data = self._memories_to_list(memories=existing_memories)
+
+        input_string = self._messages_to_input_string(messages=messages)
+
+        tools = await self._aget_extraction_tools(
+            user_id=user_id,
+            input_string=input_string,
+            existing_memories=existing_memories,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+        functions = self._build_functions_for_model(tools=tools)
+
+        messages_for_model = [
+            self._get_system_message(existing_data=existing_data),
+            *messages,
+        ]
+
+        model_copy = deepcopy(self.model)
+        response = await model_copy.aresponse(
+            messages=messages_for_model,
+            tools=functions,
+        )
+
+        if response.tool_executions:
+            self.memories_updated = True
+
+        log_debug("UserMemoryStore: Extraction complete", center=True)
+
+        return response.content or ("Memories updated" if self.memories_updated else "No updates needed")
+
+    # =========================================================================
+    # Update Operations (called by agent tool)
+    # =========================================================================
+
+    def run_memories_update(
+        self,
+        task: str,
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> str:
+        """Run a memories update task.
+
+        Args:
+            task: The update task description.
+            user_id: The unique user identifier.
+            agent_id: Agent context (stored for audit).
+            team_id: Team context (stored for audit).
+
+        Returns:
+            Response from model.
+        """
+        from agno.models.message import Message
+
+        messages = [Message(role="user", content=task)]
+        return self.extract_and_save(
+            messages=messages,
+            user_id=user_id,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+    async def arun_memories_update(
+        self,
+        task: str,
+        user_id: str,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> str:
+        """Async version of run_memories_update."""
+        from agno.models.message import Message
+
+        messages = [Message(role="user", content=task)]
+        return await self.aextract_and_save(
+            messages=messages,
+            user_id=user_id,
+            agent_id=agent_id,
+            team_id=team_id,
+        )
+
+    # =========================================================================
+    # Private Helpers
+    # =========================================================================
+
+    def _build_memories_id(self, user_id: str) -> str:
+        """Build a unique memories ID."""
+        return f"memories_{user_id}"
+
+    def _memories_to_list(self, memories: Optional[Any]) -> List[dict]:
+        """Convert memories to list of memory dicts for prompt."""
+        if not memories:
+            return []
+
+        result = []
+
+        if hasattr(memories, "memories") and memories.memories:
+            for mem in memories.memories:
+                if isinstance(mem, dict):
+                    memory_id = mem.get("id", str(uuid.uuid4())[:8])
+                    content = mem.get("content", str(mem))
+                else:
+                    memory_id = str(uuid.uuid4())[:8]
+                    content = str(mem)
+                result.append({"id": memory_id, "content": content})
+
+        return result
+
+    def _messages_to_input_string(self, messages: List["Message"]) -> str:
+        """Convert messages to input string."""
+        if len(messages) == 1:
+            return messages[0].get_content_string()
+        else:
+            return "\n".join([f"{m.role}: {m.get_content_string()}" for m in messages if m.content])
+
+    def _build_functions_for_model(self, tools: List[Callable]) -> List["Function"]:
+        """Convert callables to Functions for model."""
+        from agno.tools.function import Function
+
+        functions = []
+        seen_names = set()
+
+        for tool in tools:
+            try:
+                name = tool.__name__
+                if name in seen_names:
+                    continue
+                seen_names.add(name)
+
+                func = Function.from_callable(tool, strict=True)
+                func.strict = True
+                functions.append(func)
+                log_debug(f"Added function {func.name}")
+            except Exception as e:
+                log_warning(f"Could not add function {tool}: {e}")
+
+        return functions
+
+    def _get_system_message(
+        self,
+        existing_data: List[dict],
+    ) -> "Message":
+        """Build system message for memory extraction."""
+        from agno.models.message import Message
+
+        if self.config.system_message is not None:
+            return Message(role="system", content=self.config.system_message)
+
+        system_prompt = dedent("""\
+            You are building a memory of this user to enable personalized, contextual interactions.
+
+            Your goal is NOT to create a database of facts, but to build working knowledge that helps an AI assistant engage naturally with this person - knowing their context, adapting to their preferences, and providing continuity across conversations.
+
+            ## Memory Philosophy
+
+            Think of memories as what a thoughtful colleague would remember after working with someone:
+            - Their role and what they're working on
+            - How they prefer to communicate
+            - What matters to them and what frustrates them
+            - Ongoing projects or situations worth tracking
+
+            Memories should make future interactions feel informed and personal, not robotic or surveillance-like.
+
+            ## Memory Categories
+
+            Use memory tools for contextual information organized by relevance:
+
+            **Work/Project Context** - What they're building, their role, current focus
+            **Personal Context** - Preferences, communication style, background that shapes interactions
+            **Top of Mind** - Active situations, ongoing challenges, time-sensitive context
+            **Patterns** - How they work, what they value, recurring themes
+
+        """)
+
+        # Custom instructions or defaults
+        capture_instructions = self.config.instructions or dedent("""\
+            ## What To Capture
+
+            **DO save:**
+            - Role, company, and what they're working on
+            - Communication preferences (brevity vs detail, technical depth, tone)
+            - Goals, priorities, and current challenges
+            - Preferences that affect how to help them (tools, frameworks, approaches)
+            - Context that would be awkward to ask about again
+            - Patterns in how they think and work
+
+            **DO NOT save:**
+            - Sensitive personal information (health conditions, financial details, relationships) unless directly relevant to helping them
+            - One-off details unlikely to matter in future conversations
+            - Information they'd find creepy to have remembered
+            - Inferences or assumptions - only save what they've actually stated
+            - Duplicates of existing memories (update instead)
+            - Trivial preferences that don't affect interactions\
+        """)
+
+        system_prompt += capture_instructions
+
+        system_prompt += dedent("""
+
+            ## Writing Style
+
+            Write memories as concise, factual statements in third person:
+
+            **Good memories:**
+            - "Founder and CEO of Acme, a 10-person AI startup"
+            - "Prefers direct feedback without excessive caveats"
+            - "Currently preparing for Series A fundraise, targeting $50M"
+            - "Values simplicity over cleverness in code architecture"
+
+            **Bad memories:**
+            - "User mentioned they work at a company" (too vague)
+            - "User seems to like technology" (obvious/not useful)
+            - "Had a meeting yesterday" (not durable)
+            - "User is stressed about fundraising" (inference without direct statement)
+
+            ## Consolidation Over Accumulation
+
+            **Critical:** Prefer updating existing memories over adding new ones.
+
+            - If new information extends an existing memory, UPDATE it
+            - If new information contradicts an existing memory, REPLACE it
+            - If information is truly new and distinct, then add it
+            - Periodically consolidate related memories into cohesive summaries
+            - Delete memories that are no longer accurate or relevant
+
+            Think of memory maintenance like note-taking: a few well-organized notes beat many scattered fragments.
+
+        """)
+
+        # Current memories section
+        system_prompt += "## Current Memories\n\n"
+
+        if existing_data:
+            system_prompt += "Existing memories for this user:\n"
+            for entry in existing_data:
+                system_prompt += f"- [{entry['id']}] {entry['content']}\n"
+            system_prompt += dedent("""
+                Review these before adding new ones:
+                - UPDATE if new information extends or modifies an existing memory
+                - DELETE if a memory is no longer accurate
+                - Only ADD if the information is genuinely new and distinct
+            """)
+        else:
+            system_prompt += "No existing memories. Extract what's worth remembering from this conversation.\n"
+
+        # Available actions
+        system_prompt += "\n## Available Actions\n\n"
+
+        if self.config.enable_add_memory:
+            system_prompt += "- `add_memory`: Add a new memory (only if genuinely new information)\n"
+        if self.config.enable_update_memory:
+            system_prompt += "- `update_memory`: Update existing memory with new/corrected information\n"
+        if self.config.enable_delete_memory:
+            system_prompt += "- `delete_memory`: Remove outdated or incorrect memory\n"
+        if self.config.enable_clear_memories:
+            system_prompt += "- `clear_all_memories`: Reset all memories (use rarely)\n"
+
+        # Examples
+        system_prompt += dedent("""
+            ## Examples
+
+            **Example 1: New user introduction**
+            User: "I'm Sarah, I run engineering at Stripe. We're migrating to Kubernetes."
+            → add_memory("Engineering lead at Stripe, currently migrating infrastructure to Kubernetes")
+
+            **Example 2: Updating existing context**
+            Existing memory: "Working on Series A fundraise"
+            User: "We closed our Series A last week! $12M from Sequoia."
+            → update_memory(id, "Closed $12M Series A from Sequoia")
+
+            **Example 3: Learning preferences**
+            User: "Can you skip the explanations and just give me the code?"
+            → add_memory("Prefers concise responses with code over lengthy explanations")
+
+            **Example 4: Nothing worth saving**
+            User: "What's the weather like?"
+            → No action needed (trivial, no lasting relevance)
+
+            ## Final Guidance
+
+            - Quality over quantity: 5 great memories beat 20 mediocre ones
+            - Durability matters: save information that will still be relevant next month
+            - Respect boundaries: when in doubt about whether to save something, don't
+            - It's fine to do nothing if the conversation reveals nothing worth remembering\
+        """)
+
+        if self.config.additional_instructions:
+            system_prompt += f"\n\n{self.config.additional_instructions}"
+
+        return Message(role="system", content=system_prompt)
+
+    def _get_extraction_tools(
+        self,
+        user_id: str,
+        input_string: str,
+        existing_memories: Optional[Any] = None,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> List[Callable]:
+        """Get sync extraction tools for the model."""
+        functions: List[Callable] = []
+
+        if self.config.enable_add_memory:
+
+            def add_memory(memory: str) -> str:
+                """Save a new memory about this user.
+
+                Only add genuinely new information that will help personalize future interactions.
+                Before adding, check if this extends an existing memory (use update_memory instead).
+
+                Args:
+                    memory: Concise, factual statement in third person.
+                           Good: "Senior engineer at Stripe, working on payment infrastructure"
+                           Bad: "User works at a company" (too vague)
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    memories_data = self.get(user_id=user_id)
+                    if memories_data is None:
+                        memories_data = self.schema(user_id=user_id)
+
+                    if hasattr(memories_data, "memories"):
+                        memory_id = str(uuid.uuid4())[:8]
+                        memory_entry = {
+                            "id": memory_id,
+                            "content": memory,
+                            "source": input_string[:200] if input_string else None,
+                        }
+                        if agent_id:
+                            memory_entry["added_by_agent"] = agent_id
+                        if team_id:
+                            memory_entry["added_by_team"] = team_id
+                        memories_data.memories.append(memory_entry)
+
+                    self.save(user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id)
+                    log_debug(f"Memory added: {memory[:50]}...")
+                    return f"Memory saved: {memory}"
+                except Exception as e:
+                    log_warning(f"Error adding memory: {e}")
+                    return f"Error: {e}"
+
+            functions.append(add_memory)
+
+        if self.config.enable_update_memory:
+
+            def update_memory(memory_id: str, memory: str) -> str:
+                """Update an existing memory with new or corrected information.
+
+                Prefer updating over adding when new information extends or modifies
+                something already stored. This keeps memories consolidated and accurate.
+
+                Args:
+                    memory_id: The ID of the memory to update (shown in brackets like [abc123]).
+                    memory: The updated memory content. Should be a complete replacement,
+                           not a diff or addition.
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    memories_data = self.get(user_id=user_id)
+                    if memories_data is None:
+                        return "No memories found"
+
+                    if hasattr(memories_data, "memories"):
+                        for mem in memories_data.memories:
+                            if isinstance(mem, dict) and mem.get("id") == memory_id:
+                                mem["content"] = memory
+                                mem["source"] = input_string[:200] if input_string else None
+                                if agent_id:
+                                    mem["updated_by_agent"] = agent_id
+                                if team_id:
+                                    mem["updated_by_team"] = team_id
+                                self.save(user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id)
+                                log_debug(f"Memory updated: {memory_id}")
+                                return f"Memory updated: {memory}"
+                        return f"Memory {memory_id} not found"
+
+                    return "No memories field"
+                except Exception as e:
+                    log_warning(f"Error updating memory: {e}")
+                    return f"Error: {e}"
+
+            functions.append(update_memory)
+
+        if self.config.enable_delete_memory:
+
+            def delete_memory(memory_id: str) -> str:
+                """Remove a memory that is outdated, incorrect, or no longer relevant.
+
+                Delete when:
+                - Information is no longer accurate (e.g., they changed jobs)
+                - The memory was a misunderstanding
+                - It's been superseded by a more complete memory
+
+                Args:
+                    memory_id: The ID of the memory to delete (shown in brackets like [abc123]).
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    memories_data = self.get(user_id=user_id)
+                    if memories_data is None:
+                        return "No memories found"
+
+                    if hasattr(memories_data, "memories"):
+                        original_len = len(memories_data.memories)
+                        memories_data.memories = [
+                            mem
+                            for mem in memories_data.memories
+                            if not (isinstance(mem, dict) and mem.get("id") == memory_id)
+                        ]
+                        if len(memories_data.memories) < original_len:
+                            self.save(user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id)
+                            log_debug(f"Memory deleted: {memory_id}")
+                            return f"Memory {memory_id} deleted"
+                        return f"Memory {memory_id} not found"
+
+                    return "No memories field"
+                except Exception as e:
+                    log_warning(f"Error deleting memory: {e}")
+                    return f"Error: {e}"
+
+            functions.append(delete_memory)
+
+        if self.config.enable_clear_memories:
+
+            def clear_all_memories() -> str:
+                """Clear all memories for this user. Use sparingly.
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    self.clear(user_id=user_id, agent_id=agent_id, team_id=team_id)
+                    log_debug("All memories cleared")
+                    return "All memories cleared"
+                except Exception as e:
+                    log_warning(f"Error clearing memories: {e}")
+                    return f"Error: {e}"
+
+            functions.append(clear_all_memories)
+
+        return functions
+
+    async def _aget_extraction_tools(
+        self,
+        user_id: str,
+        input_string: str,
+        existing_memories: Optional[Any] = None,
+        agent_id: Optional[str] = None,
+        team_id: Optional[str] = None,
+    ) -> List[Callable]:
+        """Get async extraction tools for the model."""
+        functions: List[Callable] = []
+
+        if self.config.enable_add_memory:
+
+            async def add_memory(memory: str) -> str:
+                """Save a new memory about this user.
+
+                Only add genuinely new information that will help personalize future interactions.
+                Before adding, check if this extends an existing memory (use update_memory instead).
+
+                Args:
+                    memory: Concise, factual statement in third person.
+                           Good: "Senior engineer at Stripe, working on payment infrastructure"
+                           Bad: "User works at a company" (too vague)
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    memories_data = await self.aget(user_id=user_id)
+                    if memories_data is None:
+                        memories_data = self.schema(user_id=user_id)
+
+                    if hasattr(memories_data, "memories"):
+                        memory_id = str(uuid.uuid4())[:8]
+                        memory_entry = {
+                            "id": memory_id,
+                            "content": memory,
+                            "source": input_string[:200] if input_string else None,
+                        }
+                        if agent_id:
+                            memory_entry["added_by_agent"] = agent_id
+                        if team_id:
+                            memory_entry["added_by_team"] = team_id
+                        memories_data.memories.append(memory_entry)
+
+                    await self.asave(user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id)
+                    log_debug(f"Memory added: {memory[:50]}...")
+                    return f"Memory saved: {memory}"
+                except Exception as e:
+                    log_warning(f"Error adding memory: {e}")
+                    return f"Error: {e}"
+
+            functions.append(add_memory)
+
+        if self.config.enable_update_memory:
+
+            async def update_memory(memory_id: str, memory: str) -> str:
+                """Update an existing memory with new or corrected information.
+
+                Prefer updating over adding when new information extends or modifies
+                something already stored. This keeps memories consolidated and accurate.
+
+                Args:
+                    memory_id: The ID of the memory to update (shown in brackets like [abc123]).
+                    memory: The updated memory content. Should be a complete replacement,
+                           not a diff or addition.
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    memories_data = await self.aget(user_id=user_id)
+                    if memories_data is None:
+                        return "No memories found"
+
+                    if hasattr(memories_data, "memories"):
+                        for mem in memories_data.memories:
+                            if isinstance(mem, dict) and mem.get("id") == memory_id:
+                                mem["content"] = memory
+                                mem["source"] = input_string[:200] if input_string else None
+                                if agent_id:
+                                    mem["updated_by_agent"] = agent_id
+                                if team_id:
+                                    mem["updated_by_team"] = team_id
+                                await self.asave(
+                                    user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id
+                                )
+                                log_debug(f"Memory updated: {memory_id}")
+                                return f"Memory updated: {memory}"
+                        return f"Memory {memory_id} not found"
+
+                    return "No memories field"
+                except Exception as e:
+                    log_warning(f"Error updating memory: {e}")
+                    return f"Error: {e}"
+
+            functions.append(update_memory)
+
+        if self.config.enable_delete_memory:
+
+            async def delete_memory(memory_id: str) -> str:
+                """Remove a memory that is outdated, incorrect, or no longer relevant.
+
+                Delete when:
+                - Information is no longer accurate (e.g., they changed jobs)
+                - The memory was a misunderstanding
+                - It's been superseded by a more complete memory
+
+                Args:
+                    memory_id: The ID of the memory to delete (shown in brackets like [abc123]).
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    memories_data = await self.aget(user_id=user_id)
+                    if memories_data is None:
+                        return "No memories found"
+
+                    if hasattr(memories_data, "memories"):
+                        original_len = len(memories_data.memories)
+                        memories_data.memories = [
+                            mem
+                            for mem in memories_data.memories
+                            if not (isinstance(mem, dict) and mem.get("id") == memory_id)
+                        ]
+                        if len(memories_data.memories) < original_len:
+                            await self.asave(
+                                user_id=user_id, memories=memories_data, agent_id=agent_id, team_id=team_id
+                            )
+                            log_debug(f"Memory deleted: {memory_id}")
+                            return f"Memory {memory_id} deleted"
+                        return f"Memory {memory_id} not found"
+
+                    return "No memories field"
+                except Exception as e:
+                    log_warning(f"Error deleting memory: {e}")
+                    return f"Error: {e}"
+
+            functions.append(delete_memory)
+
+        if self.config.enable_clear_memories:
+
+            async def clear_all_memories() -> str:
+                """Clear all memories for this user. Use sparingly.
+
+                Returns:
+                    Confirmation message.
+                """
+                try:
+                    await self.aclear(user_id=user_id, agent_id=agent_id, team_id=team_id)
+                    log_debug("All memories cleared")
+                    return "All memories cleared"
+                except Exception as e:
+                    log_warning(f"Error clearing memories: {e}")
+                    return f"Error: {e}"
+
+            functions.append(clear_all_memories)
+
+        return functions
+
+    # =========================================================================
+    # Representation
+    # =========================================================================
+
+    def __repr__(self) -> str:
+        """String representation for debugging."""
+        has_db = self.db is not None
+        has_model = self.model is not None
+        return (
+            f"UserMemoryStore("
+            f"mode={self.config.mode.value}, "
+            f"db={has_db}, "
+            f"model={has_model}, "
+            f"enable_agent_tools={self.config.enable_agent_tools})"
+        )
+
+    def print(self, user_id: str, *, raw: bool = False) -> None:
+        """Print formatted memories.
+
+        Args:
+            user_id: The user to print memories for.
+            raw: If True, print raw dict using pprint instead of formatted panel.
+
+        Example:
+            >>> store.print(user_id="alice@example.com")
+            ╭──────────────── Memories ─────────────────╮
+            │ Memories:                                 │
+            │   [dim][a1b2c3d4][/dim] Loves Python      │
+            │   [dim][e5f6g7h8][/dim] Works at Anthropic│
+            ╰─────────────── alice@example.com ─────────╯
+        """
+        from agno.learn.utils import print_panel
+
+        memories_data = self.get(user_id=user_id)
+
+        lines = []
+
+        if memories_data:
+            if hasattr(memories_data, "memories") and memories_data.memories:
+                lines.append("Memories:")
+                for mem in memories_data.memories:
+                    if isinstance(mem, dict):
+                        mem_id = mem.get("id", "?")
+                        content = mem.get("content", str(mem))
+                    else:
+                        mem_id = "?"
+                        content = str(mem)
+                    lines.append(f"  [dim]\\[{mem_id}][/dim] {content}")
+
+        print_panel(
+            title="Memories",
+            subtitle=user_id,
+            lines=lines,
+            empty_message="No memories",
+            raw_data=memories_data,
+            raw=raw,
+        )
+
+
+# Backwards compatibility alias
+MemoriesStore = UserMemoryStore