+- **LLM Chat — talk to your timeline (RAG).** A new owner-only AI chat that answers questions over your own echos. Echos are incrementally indexed into a `sqlite-vec` vector store on create/update/delete (plus an admin full-reindex endpoint), retrieved top-k by semantic similarity, and answered with streaming SSE. Supports multi-turn conversation memory and tool-calling retrieval (`search_echos`, with tag / date filters). Embedding is configured independently via an OpenAI-compatible `/v1/embeddings` endpoint; the chat itself speaks the OpenAI or Anthropic protocol. An optional **multimodal** mode feeds matched echo images to the model, and retrieval hits surface their Extension shares (music / website / location) and image thumbnails in the UI. Entry point lives in the homepage sidebar with a dedicated `/chat` view; all settings are grouped under the Copilot panel.
0 commit comments