macOS menu bar push-to-talk speech-to-text app. 100% local on Apple Silicon.
Hold Right ⌥ → speak → release → text is pasted into the active field.
- Fully local — Voxtral Realtime 4B via MLX, no cloud, no API key
- Push-to-talk — hold Right Option to record, release to transcribe
- Live chunking — audio is transcribed during recording, not after
- Auto-paste — transcription is pasted into the active text field via Cmd+V
- macOS 14+ (Sonoma or later)
- Apple Silicon (M1/M2/M3/M4)
- ~3 GB disk space (model downloaded on first launch)
git clone https://github.com/music-mash/VoxtralPTT.git
cd VoxtralPTT
make build
xcrun swift run VoxtralPTTOr download the DMG from Releases.
On first launch, macOS will ask for:
- Microphone — to capture audio
- Accessibility — to simulate Cmd+V and paste text
| Variable | Default | Description |
|---|---|---|
VOXTRAL_MODEL |
mlx-community/Voxtral-Mini-4B-Realtime-2602-4bit |
HuggingFace model ID |
MIT