A fully offline speech-to-text app for macOS — fast, private, and entirely yours.
- On-device inference — press, speak, paste. No round-trips, no waiting.
- Zero telemetry — nothing leaves your machine. Ever.
- Minimal UI — a hotkey and a floating HUD. Stay in flow.
Default hotkey: Fn (hold) — hold to record, release to transcribe and paste.
| Mode | Behavior |
|---|---|
| Hold | Hold key to record, release to stop and paste |
| Toggle | Press to start recording, press again to stop |
Powered by Qwen3-ASR — all-in-one speech recognition supporting 30 languages and 22 Chinese dialects, with speech, singing, and background music handling.
| Model | Notes |
|---|---|
| Qwen3-ASR-0.6B | Default. Faster inference, lower memory — good accuracy-efficiency trade-off |
| Qwen3-ASR-1.7B | State-of-the-art accuracy among open-source ASR, competitive with commercial APIs |
Models are downloaded on first use.
- macOS 14+, Apple Silicon
- Microphone and Accessibility permissions (prompted on first launch)
MIT