Record, transcribe with speaker diarization, summarize, and chat across every meeting - running entirely on hardware you control. No cloud. No per-seat fees. No audio ever leaving your walls.
# one command to run the whole stack $ git clone https://github.com/kenhayward/Diariz $ cd Diariz/deploy && docker compose up -d
From raw audio to speaker-labelled transcripts, minutes, actions, and a chat assistant grounded in your own meetings.
WhisperX large-v3 with word-level timestamps and pyannote speaker diarization - editable, playable, speaker-labelled segments.
Enrol a voice once and Diariz recognises that person across every later recording - ECAPA voiceprints in pgvector.
Templated meeting minutes from reusable meeting types, an editable WYSIWYG editor, and one-click email.
Ask questions across your whole library with streaming, cited answers that link back to the exact moment.
Optional embeddings give hybrid vector + keyword search, so a concept finds the right moment even without the words.
Auto-extracted Action / Actor / Deadline, managed across every meeting, done-tracked, and folded into minutes.
Link recordings to the meeting they belong to - Google Calendar and any public iCal feed, colour-coded.
A built-in Model Context Protocol server lets Claude answer from your own transcripts - OAuth or a personal token.
Role-based access, per-user data isolation, storage quotas, backup/restore - all inside your own network.
Record from your mic, capture Windows system audio via the desktop app, or upload existing audio files.
A GPU worker runs WhisperX + pyannote and posts back speaker-labelled, timestamped segments in the background.
Read, summarize, extract actions, generate minutes, and chat or search across everything you have recorded.
Diariz runs as a handful of containers on your own machine. Transcription and diarization are fully local; summaries and chat use whatever OpenAI-compatible endpoint you point it at - including a local LLM. Ideal for regulated, sensitive, or air-gapped environments.
Read the self-host guidePersonal, academic, and non-profit use is free under AGPLv3. For-profit use without copyleft obligations needs a commercial licence.
See licensing