Local-only voice dictation for macOS. Press a hotkey, speak, the transcript appears at your cursor. Nothing leaves your device — audio, text, telemetry, nothing.
View the repository on GitHub →
brew tap larryxiao/openquack https://github.com/larryxiao/openquack
brew install --cask openquack
Or download the DMG and drag into Applications. macOS 13+ on Apple Silicon (M1 or newer).
AGENTS.md)Is OpenQuack a free alternative to Wispr Flow, SuperWhisper, or MacWhisper? Yes. OpenQuack is MIT-licensed and free. Wispr Flow, SuperWhisper, and MacWhisper require subscriptions or one-time purchase fees; OpenQuack does not.
Does it work completely offline? Yes, after the first run. On first launch, the Whisper speech model downloads from Hugging Face (~500 MB for 16 GB Macs, ~250 MB for 8 GB Macs). After that, no internet connection is needed for dictation — ever.
What Mac do I need? macOS 13 (Ventura) or later on Apple Silicon (M1 or newer). Intel Macs are not supported.
How accurate is it? On real human speech, ~2.6% word-error rate with the default model on an M4 / 16 GB Mac. In realistic office noise: ~6.3% WER. Full benchmark matrix in BENCHMARKS.md.
Does it send my audio or transcripts anywhere? No. Audio is recorded, transcribed, and discarded entirely on your Mac. No analytics, no telemetry, no account, no API calls in the dictation path. Source is MIT-licensed and auditable.
How is this different from the built-in macOS Dictation? macOS Dictation sends audio to Apple’s servers by default. OpenQuack runs fully local, supports 99 languages with no toggle, uses Whisper rather than Apple’s proprietary model, and is open source.
How do I use it for typeless coding workflows? OpenQuack pastes at wherever your cursor is — including the prompt bars of Claude Code, Cursor, Windsurf, or any terminal. Press the hotkey, speak the prompt, press again, and it appears.
Is there a Windows or Linux version? Not currently. OpenQuack uses WhisperKit and CoreML, which are Apple-platform technologies.
MIT. See LICENSE in the repository.