AmicalLocal AI Models
Run AI models locally on your device for private, offline transcription
Amical supports running Whisper speech-to-text models entirely on your device. No internet required, no data leaves your machine.
Available Models
Download models in Settings → AI Models → Speech. Each model shows speed and accuracy ratings to help you choose.
| Model | Size | Best For |
|---|---|---|
| Tiny | ~78 MB | Fastest transcription, lower accuracy |
| Base | ~148 MB | Balanced speed and accuracy |
| Small | ~488 MB | Good accuracy for most use cases |
| Medium | ~1.5 GB | High accuracy |
| Large v3 | ~3.1 GB | Highest accuracy |
| Large v3 Turbo | ~1.5 GB | Optimized — near Large v3 accuracy at faster speed |
Setup
- Open Settings → AI Models → Speech
- Click the download button next to your preferred model
- Wait for the download to complete
- Select the model via the radio button to activate it
You can download multiple models and switch between them at any time.
Text Formatting Models
For the AI formatting step (filler removal, grammar, context-aware formatting), Amical supports:
- Ollama — Run local LLMs on your machine for fully offline formatting
- OpenRouter — Access cloud LLMs with your own API key
Configure formatting models in Settings → AI Models.