What TransMeet sees, and what it doesn't.
The full policy ships alongside v0.1.0. The promises below are what the engine already enforces today.
Audio never leaves your device. Capture, voice activity detection, transcription (faster-whisper), and speaker diarization (ECAPA-TDNN) all run locally. There is no remote audio fallback, and there never will be.
Transcripts stay local by default. Every meeting writes a JSONL file under ~/.transmeet/sessions/. You decide if and when to export them.
LLM features are opt-in. Summary, action items, and the Co-Pilot only run when you have provided a Gemini API key. Without a key the relevant endpoints return 503 LLM_NOT_CONFIGURED. We never proxy through a TransMeet-owned key.
Voiceprints are local. Speaker embeddings live in the same per-machine SQLite database as your sessions. They are never uploaded.
No analytics ping the network. The engine does not phone home. The Chrome extension talks only to your local engine on 127.0.0.1:8765.