Docs
Chat & Search
Two ways to get things back out of your memory: hybrid search across the timeline, and a conversational agent that answers from your history.
See chat in action
Hybrid search
Search fuses two signals: semantic embeddings (find by meaning) and an FTS5 keyword index (find exact terms), plus meeting-transcript matches. Results are merged and re-ranked, and matching text is highlighted right on the screenshot.
Chat with your memory
Chat behaves like a normal assistant that can reach into your timeline when relevant:
1
Keyword probe
Keywords are pulled from your question and run through FTS5 to see if the timeline has anything relevant.
2
Semantic re-rank
Candidate moments are re-scored with embeddings, plus a gentle recency boost, to pick the best context.
3
Answer
The organized screen text (or the screenshot itself, as a vision fallback) is handed to Gemma to answer — with conversation history for follow-ups.
Casual chit-chat skips the timeline lookup entirely, and chat always pre-empts background analysis so answers come fast. Responses stream over SSE.
Ask things like “what did Alex say on Discord earlier?” or “what was I working on at 3pm?” — it pulls the actual moment, not a guess.
↯ back to the brain