Diariz
Self-hosted meeting transcription with local audio processing, speaker recognition and chat through local or cloud OpenAI-compatible models. AGPLv3.
Diariz is a self-hosted meeting transcription app for individuals and teams that want recordings and transcripts on their own server. It runs in Docker, with a browser interface and desktop apps for Windows and macOS (beta) for microphone and system audio capture. You can also upload existing recordings. Its source is available under AGPLv3.
Transcription and speaker diarization run locally on CPU or GPU using WhisperX large-v3 and pyannote. Transcripts have word-level timestamps and editable, playable segments labelled by speaker. After you enrol someone's voice, Diariz can recognise that person in later recordings, so recurring meetings retain named speakers.
Summaries and chat use an OpenAI-compatible endpoint, including local models served by Ollama, LM Studio or vLLM. Local processing can keep the workflow on your hardware; choosing a cloud endpoint sends the content needed for those requests to that service. The built-in MCP server also lets Claude access your transcripts.
Across your meeting library, chat answers cite the exact moment they draw on. Search combines keywords with optional semantic embeddings. Diariz extracts action items with an owner and deadline, tracks completion, and includes them in editable minutes based on reusable meeting templates. Google Calendar and public iCal feeds link recordings to events, while recurring meetings show earlier recordings. For shared servers, role-based access and per-user data isolation control access to meeting records.
