Add user memory feature for cross-conversation recall

Give the LangChain agent a save_user_memory tool so users can ask it to
remember preferences and personal facts. Memories are stored per-user in
a new user_memories table and injected into the system prompt on each
conversation turn.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

blueprints/conversation/__init__.py

@@ -22,6 +22,7 @@ from .logic import (
     get_conversation_by_id,
     rename_conversation,
 )
+from .memory import get_memories_for_user
 from .models import (
     Conversation,
     PydConversation,
@@ -36,15 +37,27 @@ conversation_blueprint = Blueprint(
 _SYSTEM_PROMPT = SIMBA_SYSTEM_PROMPT
 
 
+async def _build_system_prompt_with_memories(user_id: str) -> str:
+    """Append user memories to the base system prompt."""
+    memories = await get_memories_for_user(user_id)
+    if not memories:
+        return _SYSTEM_PROMPT
+    memory_block = "\n".join(f"- {m}" for m in memories)
+    return f"{_SYSTEM_PROMPT}\n\nUSER MEMORIES (facts the user has asked you to remember):\n{memory_block}"
+
+
 def _build_messages_payload(
-    conversation, query_text: str, image_description: str | None = None
+    conversation,
+    query_text: str,
+    image_description: str | None = None,
+    system_prompt: str | None = None,
 ) -> list:
     recent_messages = (
         conversation.messages[-10:]
         if len(conversation.messages) > 10
         else conversation.messages
     )
-    messages_payload = [{"role": "system", "content": _SYSTEM_PROMPT}]
+    messages_payload = [{"role": "system", "content": system_prompt or _SYSTEM_PROMPT}]
     for msg in recent_messages[:-1]:  # Exclude the message we just added
         role = "user" if msg.speaker == "user" else "assistant"
         text = msg.text
@@ -80,10 +93,14 @@ async def query():
         user=user,
     )
 
-    messages_payload = _build_messages_payload(conversation, query)
+    system_prompt = await _build_system_prompt_with_memories(str(user.id))
+    messages_payload = _build_messages_payload(
+        conversation, query, system_prompt=system_prompt
+    )
     payload = {"messages": messages_payload}
+    agent_config = {"configurable": {"user_id": str(user.id)}}
 
-    response = await main_agent.ainvoke(payload)
+    response = await main_agent.ainvoke(payload, config=agent_config)
     message = response.get("messages", [])[-1].content
     await add_message_to_conversation(
         conversation=conversation,
@@ -163,15 +180,19 @@ async def stream_query():
             logging.error(f"Failed to analyze image: {e}")
             image_description = "[Image could not be analyzed]"
 
+    system_prompt = await _build_system_prompt_with_memories(str(user.id))
     messages_payload = _build_messages_payload(
-        conversation, query_text or "", image_description
+        conversation, query_text or "", image_description, system_prompt=system_prompt
     )
     payload = {"messages": messages_payload}
+    agent_config = {"configurable": {"user_id": str(user.id)}}
 
     async def event_generator():
         final_message = None
         try:
-            async for event in main_agent.astream_events(payload, version="v2"):
+            async for event in main_agent.astream_events(
+                payload, version="v2", config=agent_config
+            ):
                 event_type = event.get("event")
 
                 if event_type == "on_tool_start":

blueprints/conversation/agents.py

@@ -5,9 +5,11 @@ from dotenv import load_dotenv
 from langchain.agents import create_agent
 from langchain.chat_models import BaseChatModel
 from langchain.tools import tool
+from langchain_core.runnables import RunnableConfig
 from langchain_openai import ChatOpenAI
 from tavily import AsyncTavilyClient
 
+from blueprints.conversation.memory import save_memory
 from blueprints.rag.logic import query_vector_store
 from utils.obsidian_service import ObsidianService
 from utils.ynab_service import YNABService
@@ -589,8 +591,35 @@ async def obsidian_create_task(
         return f"Error creating task: {str(e)}"
 
 
+@tool
+async def save_user_memory(content: str, config: RunnableConfig) -> str:
+    """Save a fact or preference about the user for future conversations.
+
+    Use this tool when the user:
+    - Explicitly asks you to remember something ("remember that...", "keep in mind...")
+    - Shares a personal preference that would be useful in future conversations
+      (e.g., "I prefer metric units", "my cat's name is Luna")
+    - Tells you a meaningful personal fact (e.g., "I'm allergic to peanuts")
+
+    Do NOT save:
+    - Trivial or ephemeral info (e.g., "I'm tired today")
+    - Information already in the system prompt or documents
+    - Conversation-specific context that won't matter later
+
+    Args:
+        content: A concise statement of the fact or preference to remember.
+                 Write it as a standalone sentence (e.g., "User prefers dark mode"
+                 rather than "likes dark mode").
+
+    Returns:
+        Confirmation that the memory was saved.
+    """
+    user_id = config["configurable"]["user_id"]
+    return await save_memory(user_id=user_id, content=content)
+
+
 # Create tools list based on what's available
-tools = [get_current_date, simba_search, web_search]
+tools = [get_current_date, simba_search, web_search, save_user_memory]
 if ynab_enabled:
     tools.extend(
         [

blueprints/conversation/memory.py

@@ -0,0 +1,19 @@
+from .models import UserMemory
+
+
+async def get_memories_for_user(user_id: str) -> list[str]:
+    """Return all memory content strings for a user, ordered by most recently updated."""
+    memories = await UserMemory.filter(user_id=user_id).order_by("-updated_at")
+    return [m.content for m in memories]
+
+
+async def save_memory(user_id: str, content: str) -> str:
+    """Save a new memory or touch an existing one (exact-match dedup)."""
+    existing = await UserMemory.filter(user_id=user_id, content=content).first()
+    if existing:
+        existing.updated_at = None  # auto_now=True will refresh it on save
+        await existing.save(update_fields=["updated_at"])
+        return "Memory already exists (refreshed)."
+
+    await UserMemory.create(user_id=user_id, content=content)
+    return "Memory saved."

blueprints/conversation/models.py

@@ -47,6 +47,17 @@ class ConversationMessage(Model):
         table = "conversation_messages"
 
 
+class UserMemory(Model):
+    id = fields.UUIDField(primary_key=True)
+    user = fields.ForeignKeyField("models.User", related_name="memories")
+    content = fields.TextField()
+    created_at = fields.DatetimeField(auto_now_add=True)
+    updated_at = fields.DatetimeField(auto_now=True)
+
+    class Meta:
+        table = "user_memories"
+
+
 PydConversationMessage = pydantic_model_creator(ConversationMessage)
 PydConversation = pydantic_model_creator(
     Conversation, name="Conversation", allow_cycles=True, exclude=("user",)

blueprints/conversation/prompts.py

@@ -54,4 +54,7 @@ You have access to Ryan's daily journal notes. Each note lives at journal/YYYY/Y
 - Use journal_get_tasks to list tasks (done/pending) for today or a specific date
 - Use journal_add_task to add a new task to today's (or a given date's) note
 - Use journal_complete_task to check off a task as done
-Use these tools when Ryan asks about today's tasks, wants to add something to his list, or wants to mark a task complete."""
+Use these tools when Ryan asks about today's tasks, wants to add something to his list, or wants to mark a task complete.
+
+USER MEMORY:
+You can remember facts about the user across conversations using the save_user_memory tool. When a user explicitly asks you to remember something, or shares a meaningful preference or personal fact, save it. Saved memories will automatically appear at the end of this prompt in future conversations under "USER MEMORIES"."""