Press. Speak. Done. Your voice belongs on your Mac.
Press the right ⌘ key, speak naturally, press it again, and your words are transcribed on-device and pasted into whatever app you're using.
No cloud. No account. No subscriptions. No word limits.
1. Press to record — a native CGEventTap catches a quick tap of the right ⌘ key. The menu-bar icon turns red.
2. Speak — malgo captures your mic at 16 kHz, fully on-device.
3. Press again — whisper.cpp transcribes locally with Metal acceleration, over 30× faster than real time.
4. Done — your text is pasted into the active app and copied to your clipboard.
Nothing leaves your Mac. The right ⌘ key still works normally as a modifier in keyboard shortcuts — only a quick standalone tap triggers dictation.
🔒 Private by design — every transcription happens on your Mac. Your voice never leaves your device.
💸 Free forever — open source, no subscriptions, no API keys, no limits.
⚡ Blazing fast — Metal-accelerated whisper.cpp on Apple Silicon transcribes faster than you can re-read it.
🌍 Multilingual — auto-detects the language you speak and writes it back in that language.
🪶 Invisible — a tiny menu-bar app. Zero onboarding, zero accounts, zero distractions.
curl -fsSL https://raw.githubusercontent.com/kzkr/voysnap/main/install.sh | bashThat's it — no Xcode, no Go, no build step. The installer drops VoySnap.app into /Applications, downloads the speech model (~1.5 GB) on first run, and launches it.
On first launch, macOS asks for two permissions:
- 🎤 Microphone — to hear you.
- ♿ Accessibility — for the global hotkey and automatic pasting.
That's it.
Build from source
Requires macOS 13+ · Apple Silicon · Homebrew · Go 1.24+ · CMake · Xcode Command Line Tools (xcode-select --install).
git clone git@github.com:kzkr/voysnap.git
cd voysnap
make installmake install builds whisper.cpp, downloads the model, signs the app, and installs it into /Applications. The first build takes a few minutes; after that it's nearly instant.
- Put your cursor where you want to type.
- Tap the right ⌘ key — the menu-bar icon turns red.
- Speak.
- Tap right ⌘ again — your text appears.
Text is pasted exactly as Whisper transcribes it, including punctuation and capitalization. When nothing editable is focused — e.g. on the desktop — VoySnap shows the result in a popup and leaves it on your clipboard instead.
VoySnap works out of the box. To customize it, edit ~/Library/Application Support/VoySnap/config.json:
| Key | Description |
|---|---|
language |
"auto" (default), or a code such as "en" / "fr" |
model_path |
path to a different whisper model |
vocabulary |
custom words, names, or jargon to recognize better |
snippets |
{ "spoken phrase": "replacement" } text expansions |
All transcription runs locally via whisper.cpp. See CLAUDE.md for architecture and design notes.