Models & licenses
Which models Zerds runs, and the license terms that come with them.
Models download on first use from their providers over HTTPS, are cached locally, and run offline afterward. No audio or personal data is sent during a download — it's a plain file fetch. Sizes are listed under Transcription & models.
Transcription models
| Model | License | Note |
|---|---|---|
| OpenAI Whisper | MIT | Default engine; 99+ languages |
| NVIDIA Parakeet TDT | CC-BY-4.0 | Optional; attribution required; 25 languages |
Local AI models (optional)
If you use the built-in AI, you pick one of several models — each with its own license:
| Model | License |
|---|---|
| Qwen2.5 / Qwen3 | Apache-2.0 |
| Gemma 2 / Gemma 3 | Gemma Terms of Use |
| Llama 3.2 | Llama Community License |
| Phi-3.5 mini | MIT |
The app ships with a NOTICES file containing the full terms. A few models (notably Llama and Gemma) carry attribution or acceptable-use conditions you'd need to honor if you redistribute — for personal use in Zerds, they just run locally.