Free on-device dictation. Speak, and text appears in whatever app you are using. Audio never leaves your device: no servers, no account, no API keys, no cost.
Download the latest release · Website
Each desktop OS has its own dedicated tree and build. They do not share an engine.
| Platform | Code | Build | Engine |
|---|---|---|---|
| macOS | mac/ |
cd mac && ./build.sh (Swift / AppKit) |
MLX + llama.cpp + whisper.cpp + Apple Speech |
| Windows | windows/Scribe/ |
dotnet build -c Release (.NET 8 WinForms) |
sherpa-onnx, GPU auto (CUDA / DirectML / CPU) |
| Android | android/ + src/ |
npx expo run:android |
mobile ASR stack |
| iOS | ios/ + src/ |
npx expo run:ios |
mobile ASR stack |
There is no Linux desktop app yet. Linux would get its own folder, not a port of mac/ or windows/.
- Dictation into any app. Hold a hotkey, speak, release. Text lands at your cursor.
- Your choice of model. Whisper, Parakeet, Moonshine, Canary, Dolphin, Nemotron, and Srota (Hinglish). Download the ones you want, each tagged by size and quality.
- Hinglish. Hindi comes out in English letters ("ab isko badalne ke liye"), the way people actually type it. Toggleable.
- Spoken commands. "next line", "new paragraph", "point one" for numbered lists, "bullet" for dashes, "scratch that" to undo a line.
- AI cleanup and summary. Optional on-device Gemma model polishes the transcript. Also fully local.
- Light on memory. Models load on demand and unload after 5 minutes idle.
macOS (Swift, menu bar app):
cd mac && ./build.sh
open Scribe.app # grant Mic, Speech, and AccessibilityMobile (Expo, native build required, does not run in Expo Go):
npx expo run:ios # or run:androidWindows (C# tray app, separate from Mac):
cd windows\Scribe
dotnet build -c Release
dotnet publish -c Release -r win-x64 --self-contained -p:PublishSingleFile=trueEverything runs on your device. No telemetry, no network calls except model downloads. See the privacy policy.
MIT