A fast, native, privacy-first dictation app for macOS and Windows. On-device transcription with WhisperKit — no account, no cloud, no API key required.
Download · Features · How it works · Build from source · Contributing
You think at ~150 words per minute. You type at ~40. AIT-Scribe closes that gap everywhere on your computer — your editor, your browser, Slack, email, a terminal — without ever leaving the app you're in.
Press and hold one key, talk naturally, release. The text is transcribed and pasted at your cursor. That's the whole interaction.
Privacy by default. The flagship engine runs Whisper entirely on your device. Your voice never leaves your machine, it works on a plane with no Wi-Fi, and it costs nothing per word. Cloud engines are strictly opt-in.
| AIT-Scribe | |
|---|---|
| 🖥️ Native, not Electron | Swift + SwiftUI on macOS, .NET 8 + WinUI 3 on Windows. Low memory, instant launch. |
| 🔒 On-device transcription | WhisperKit / Whisper runs locally. Offline-capable, no API key, no per-word cost. |
| 🌍 Works in every app | A global hotkey + auto-paste means dictation works system-wide — no integrations needed. |
| 🗣️ Multiple engines | Local Whisper, OpenAI, or Sarvam (great for Indian languages). Pick per your needs. |
| 📚 Learns your words | Custom vocabulary + auto-learn capture names, jargon, and acronyms you actually use. |
| 🪟 Cross-platform | One project, two first-class native apps. |
- Push-to-talk dictation — hold
Fn(macOS) orRight Ctrl(Windows), speak, release. Text lands at your cursor. - On-device Whisper — choose a model size (
base→turbo→large-v3) to balance speed and accuracy. First run downloads the model; everything after is offline. - Cloud engines (optional) — OpenAI and Sarvam transcription for when you want them. Opt-in, never default.
- Floating status bubble — a small, draggable widget shows recording state and works over fullscreen apps.
- Custom vocabulary — teach it your names, brands, and technical terms so they're spelled right every time.
- Auto-learn — when you correct a single word after pasting, AIT-Scribe offers to remember it.
- Smart cleanup — built-in filler-word, hallucination, and profanity filters keep transcripts tidy.
- Native menu bar / system tray — stays out of your way until you need it.
- In-app updates — the app checks for new releases and updates itself.
📸 Screenshots & demo GIF: drop them into
docs/and reference them here — see docs/README.md for the recommended shots (hero, bubble while recording, settings, vocabulary). Real screenshots are the single biggest driver of stars; add them before announcing.
Pre-built, signed binaries are available from scribe.aithinkers.com.
| Platform | File | Requirements |
|---|---|---|
| macOS | AIT-Scribe-macOS.dmg |
macOS 14 (Sonoma) or later, Apple Silicon or Intel |
| Windows | AIT-Scribe-Windows-Setup.exe |
Windows 10 19041+ / Windows 11 |
macOS first launch: if Gatekeeper blocks the app, open System Settings → Privacy & Security, scroll down, and click Open Anyway.
Prefer to build it yourself? See Build from source.
┌── Hold hotkey ──┐ ┌── Transcribe ──┐ ┌── Output ──┐
│ Fn / RCtrl │ ───▶ │ Whisper (local)│ ───▶ │ Auto-paste │
│ starts capture │ │ OpenAI/Sarvam │ │ at cursor │
└─────────────────┘ └────────────────┘ └────────────┘
AudioRecorder TranscriptionService ClipboardManager
(mic input) (engine + vocab + filters) (simulated ⌘V / Ctrl+V)
- Capture — a global key listener (Quartz event tap on macOS, low-level hook on Windows) records mic audio while the hotkey is held.
- Transcribe — audio is sent to the selected engine. The local engine runs Whisper on-device via WhisperKit; cloud engines call their APIs only if you've enabled them.
- Refine — your custom vocabulary is applied and filler/hallucination/profanity filters clean the result.
- Output — the text is placed on the clipboard and pasted at your cursor with a simulated paste keystroke.
Requires macOS 14+, Xcode 15+ (from the Mac App Store), and Homebrew.
git clone https://github.com/aithinkers/scribe.git
cd scribe/"AIT-Scribe(Mac)"
# Generates the Xcode project (installs XcodeGen via brew if needed) and opens it
./setup.sh
# …or build an unsigned dev build straight to disk
./setup.sh buildThe first launch downloads a Whisper model to ~/Documents/huggingface/ (multi-GB for larger models — base is a fast starting point). Grant the permissions the app requests (see below).
AIT-Scribe needs three permissions, all granted in System Settings → Privacy & Security:
| Permission | Why |
|---|---|
| Microphone | Record your voice while the hotkey is held |
| Input Monitoring | Detect the global hotkey press/release |
| Accessibility | Paste transcribed text at your cursor (simulated ⌘V) |
After granting them, restart the app.
Requires .NET 8 SDK and the Windows App SDK workload. Visual Studio 2022 (17.8+) with the .NET Desktop + WinUI workloads is the easiest path.
git clone https://github.com/aithinkers/scribe.git
cd scribe/AIT-Scribe-Windows
# Restore + build
dotnet restore
dotnet build AIT-Scribe.sln -c Release
# Run
dotnet run --project AIT.Scribe.Desktop -c ReleaseOr open AIT-Scribe-Windows/AIT-Scribe.sln in Visual Studio and press F5.
On-device transcription needs no configuration. To enable cloud engines or account sync, copy .env.example to .env and fill in your own backend/Auth0/Sarvam values:
cp .env.example .envThe repository ships with placeholder endpoints (your-backend.example.com, your-tenant.us.auth0.com) — bring your own backend if you want the cloud features. The local dictation path works fully without any of them.
Contributions are very welcome — bug reports, fixes, docs, and features. Start with CONTRIBUTING.md and look for good first issue labels.
If AIT-Scribe is useful to you, the best way to help is to ⭐ star the repo — it genuinely helps other people find it.
- WhisperKit — on-device Whisper inference for Apple platforms
- OpenAI Whisper — the speech-recognition model family
- Windows App SDK / WinUI 3 — native Windows UI
- XcodeGen — Xcode project generation
MIT © AI Thinkers LLC. Use it, fork it, ship it.