Licensing

Free to self-host. Commercial when you need it.

Diariz is dual-licensed. Choose the option that fits your use.

Open Source

GNU AGPLv3 - free forever
  • ✓ Personal, academic, and non-profit use
  • ✓ Full source, self-hosted, no seat limits
  • ✓ Modify and extend under AGPLv3

Commercial

Private licence for for-profit use
  • ✓ Use without AGPLv3 copyleft obligations
  • ✓ Custom terms and exceptions
  • ✓ Talk to us: ken@stocks-hayward.com

Before you request a commercial licence

A few parts of the ML/storage stack carry caveats worth understanding. This is a summary for orientation, not legal advice.

Transcription & diarization

Whisper large-v3 (MIT) and the pyannote models are MIT-licensed. They are gated on Hugging Face (accept the terms, supply an HF token) - gating is an access step, not a licence restriction. Clear for commercial use.

Speaker identification / voiceprints

Cross-recording speaker ID uses SpeechBrain ECAPA embeddings. The code is Apache-2.0 but the weights are trained on VoxCeleb (research / non-commercial). For commercial use, get your own legal read, swap the embedder for commercially-cleared weights, or disable the feature. Voiceprints are biometric data - enrol only with consent.

Object storage (MinIO)

MinIO is AGPL-3.0. Used unmodified as a separate container it does not impose copyleft on Diariz’s own code; if AGPL is a concern, point storage at any S3-compatible store instead.

Summaries & chat

These send transcript text to whatever OpenAI-compatible endpoint you configure; that provider’s terms govern the text you send. Point it at a local LLM to keep everything in-house.

Uploaded audio formats

Decoding is ffmpeg in the worker. Royalty-free formats (WAV, FLAC, Ogg, Opus, WebM) plus MP3 always pass; M4A/AAC still carries patents and can be disabled for maximum caution.

To discuss commercial licensing, custom terms, or an exception, contact ken@stocks-hayward.com.