semantic-history-search

You can search past conversations by meaning using semantic embeddings combined with BM25 keyword search. This hybrid approach merges results via Reciprocal Rank Fusion to retrieve relevant context based on intent rather than exact wording.

Cross-session recall allows you to retrieve context from multiple conversations and channels. It indexes past interactions into embedded memory chunks, enabling topic tracking and decision provenance across separate sessions.

You need the user/embeddings.db database for storage and context. Memory-backed results rely on this database to persist embedded memory chunks, which are then queried during the hybrid semantic and keyword retrieval process.

Reciprocal Rank Fusion improves conversation retrieval by merging semantic embeddings with BM25 keyword search results. This combines meaning-based matching with exact term frequency to surface the most relevant past conversations.

The best way to track decisions across chat histories is using hybrid retrieval over embedded memory chunks. By applying semantic embeddings and keyword search, you can trace decision provenance and track topics across multiple sessions.

A limitation of semantic search for session history is its dependence on the embeddings.db storage. If past conversations are not indexed into embedded memory chunks, cross-session recall and topic tracking will not return results.