Skip to main content

Conversation History

The memory manager supports storing and retrieving conversation messages using a ConversationID.

Local Runtime: auto-managed history​

If you are using the LocalRuntime (default), you generally do not need to store messages yourself. The runtime automatically:

  • stores each incoming message in the conversation store
  • attaches recent conversation_history to the AgentContext
  • provides a history_provider on AgentContext (used by PromptBuilder when you set k_limit)

Only store messages manually if you are building a custom runtime or invoking agents directly without going through the runtime.

View the source code.

Store Messages​

from l6e_forge.types.core import Message, AgentContext

async def handle_message(self, message: Message, context: AgentContext):
    mm = self.runtime.get_memory_manager()
    # Persist the latest user message
    await mm.store_conversation(context.conversation_id, message)

Note: When using LocalRuntime.route_message(...), this store step is already performed for you.

Read Recent History​

history = await mm.get_conversation(context.conversation_id, limit=50)

When a conversation store is configured (e.g., Postgres), messages are persisted there; otherwise an in-memory store is used for the session.

You can combine conversation history with vector search to retrieve relevant context for prompting.