Memory Retrieval

To preserve and retrieve context across sessions, you index past conversation text with metadata like session_id and timestamp. This enables fast cross-session similarity search to recall relevant past information for current prompts.

Semantic memory indexing stores past context with metadata for later retrieval. You need it when conversations get voluminous or span multiple sessions, requiring similarity search to find relevant past context to support current prompts.

You search past conversations by applying similarity search alongside deterministic filtering by metadata. The system indexes text with session_id, timestamp, and turn information, enabling precise retrieval of specific past interactions.

Yes, you need a MemoryStore backend to handle the storage and retrieval of memory items. The system exposes indexing and search APIs that interact with this backend to deterministically store, retrieve, and filter memories by metadata.

The best way to recall a user preference is to index the text containing the preference with session metadata, then perform a cross-session similarity search to retrieve that specific context and tailor your current response.

Yes, this approach is designed for large volumes of conversations. By indexing text with metadata and utilizing semantic search, you can quickly find relevant past context to support current prompts even when the conversation history is extensive.