AtomVoice
Privacy-first voice dictation for Mac
Tap or hold a key, speak naturally, and your words appear.
macOS 14+ · Free · Open Source
Privacy First
Speech recognition runs on-device by default. No audio leaves your Mac unless you explicitly choose a cloud engine.
Lightweight
Small bundle, near-zero CPU when idle. No background daemons, no telemetry, no accounts.
Recognition Engines
Apple Speech
Default
Built-in, streaming. Optional on-device mode. No time limit thanks to segmented rolling recognition.
Sherpa-ONNX
Offline
Fully offline. Models download automatically on first use, released under memory pressure.
Volcengine
Cloud
Volcengine (Doubao) real-time streaming cloud ASR via Bytedance. Requires API key configuration.
More Features
- Hold-to-talk or tap-to-talk
- No recording time limit
- Optional LLM refinement
- 8 recognition languages
- CJK IME compatible
- Live waveform animation
- Auto-update from GitHub
- Launch at login
Download
Universal, Apple Silicon, and Intel builds — all on GitHub Releases.
Or install via Homebrew
brew tap BlackSquarre/tap
brew install --cask atomvoice