Press a key. Speak. The transcript appears at your cursor — in any app.
Local-first speech-to-text for macOS. No cloud. No account. No telemetry.
System dictation on macOS sends your voice to Apple's servers. Most local STT tools require a terminal and Python skills.
WhisperLocal is the alternative: a tiny menu bar app, fully offline, that just works. Hold a hotkey or tap it, speak, and your words land at the cursor — in Terminal, VS Code, Slack, Notes, the address bar of your browser, anywhere.
- 🎙 System-wide dictation — works in every text field on your Mac
- ⌨️ Tap to start, stay silent to stop — or push-to-talk if you prefer
- 🇩🇪🇬🇧 German + English with auto-detection
- 🧠 Apple Intelligence cleanup removes "ähm", fixes punctuation (optional)
- 🔒 100 % local — nothing ever leaves your Mac
- ⚡️ Sub-second latency on Apple Silicon thanks to CoreML
- 🎨 Looks like a Mac app should — menu bar icon, floating overlay, native settings
- Grab the latest WhisperLocal-x.y.z.zip from Releases.
- Unzip and drag
WhisperLocal.appinto/Applications. - Open it. Grant Microphone, Accessibility, and Input Monitoring when prompted. Relaunch once after granting.
The release is signed with a Developer ID certificate and notarized by Apple, so Gatekeeper opens it without warnings.
git clone https://github.com/marco-scheffler/WhisperLocal.git
cd WhisperLocal
./Scripts/create-dev-cert.sh # one-time
./Scripts/install.shSee docs/DEVELOPMENT.md for the full guide.
- Press and hold (or tap) ⌥⌘ — the default hotkey
- Speak
- Stop by tapping again, or just stay silent for 2 seconds
That's it. The cleaned transcript is pasted wherever your cursor is.
Every setting — hotkey, language, microphone, post-processing, autostart — lives in the menu bar icon's Settings… window.
- Whisper runs locally via WhisperKit (CoreML)
- LLM cleanup uses Apple's on-device FoundationModels
- The Whisper model is downloaded once from HuggingFace; everything else stays on your Mac
- No analytics, no crash reporting, no third-party SDKs
- Read the privacy section in the docs for details
- macOS 15 (Sequoia) or later — macOS 26 (Tahoe) recommended
- Apple Silicon (M1 or newer)
- ~600 MB free disk for the Whisper model
- 📖 Developer & maintainer docs — architecture, all settings, troubleshooting, release process
- 🐛 Issues — bug reports & ideas welcome
- 📄 License — MIT
Built with WhisperKit, OpenAI Whisper, Apple FoundationModels, and Swift 6.