tapes-search

To perform semantic search over past LLM sessions, you run a query against the stored session data. The skill uses an embedding provider to convert your query into a vector and surfaces relevant conversations from the historical telemetry data.

To retrieve context from historical LLM interactions, you need a running vector store like Chroma with indexed session data, an embedding provider like Ollama for query vectorization, and a SQLite database containing the stored session telemetry.

Yes, you can locate specific discussions or topics within stored LLM telemetry. It indexes past interactions, allowing you to search through historical data and recall exact context from previous sessions based on semantic similarity.

Yes, semantic search for past conversations works with SQLite and Chroma. SQLite stores the session data, while Chroma acts as the vector store to index embeddings, enabling the retrieval of relevant past interactions.

An embedding provider like Ollama converts your textual search query into a numerical vector. This vector is then compared against indexed session data in the vector store to find and surface semantically relevant past LLM conversations.

The main limitation is the dependency on external infrastructure; the semantic search requires a pre-configured vector store, an active embedding provider, and a populated SQLite database to successfully surface relevant past conversations.