Type at the speed of thought. Fully offline. Fully private.
Download · Features · Showcase · Compare · Docs · Contribute
Built with Tauri v2 (Rust) + React — Dictate anywhere, your words appear instantly in any application.
|
All transcription runs locally on your device by default — no audio ever leaves your machine unless you explicitly enable cloud transcription. No accounts, no subscriptions. Your voice stays yours. |
Powered by NVIDIA Parakeet, Whisper, and Moonshine engines (via transcribe-rs) for blazing-fast, real-time voice-to-text. Start speaking and see words appear instantly. |
|
Optional LLM post-processing cleans up filler words, fixes grammar, and formats your text — via cloud providers (OpenAI-compatible, Groq, and others). |
Transcribe in 50+ languages with automatic language detection. Offline, pick your engine: Parakeet V3 auto-detects languages, Whisper covers 99 languages, and cloud Whisper extends coverage further. Switch languages on the fly or lock to a specific one. |
|
SONU types directly into any application — your browser, IDE, email client, Slack, Discord, Word — anywhere you can type. |
SONU detects the app you're typing in and adapts tone and formatting automatically — casual in messengers, formal in email, code-safe in your IDE. |
|
Select any text, press the Command Mode shortcut, and speak an instruction like "make this more concise" — your voice-driven AI rewrites it in place. |
Connect to Groq, Deepgram, or any custom API endpoint for cloud-powered transcription when you want maximum accuracy. |
|
Custom word corrections automatically fix domain-specific terms, names, and jargon that the model might mishear. |
Define shorthand codes that expand into full text blocks — perfect for emails, code comments, addresses, and common phrases. |
SONU v2.6.0 — Built with Tauri v2 (Rust + React). Lightweight, native, and fast.
SONU ships with bundled Geist typography, light/dark/system themes, six accent presets, semantic design tokens, and a low-latency floating recording overlay. The local live preview begins showing text while you speak; the final pass keeps full accuracy and existing AI enhancement behavior.
The screenshot above is captured from the Windows Tauri v2 app. More theme screenshots will be added as part of the cross-platform visual QA pass.
The home screen shows your dictation stats (time, word count, WPM, time saved), a voice activation shortcut recorder, privacy status, and recent transcription history — all in a clean dashboard layout with local/cloud mode indicator.
Dictionary lets you add custom word corrections for domain-specific terms the model might mishear. Snippets are reusable text blocks you can expand with shorthand codes — perfect for emails, addresses, and common phrases.
Voice-powered sticky notes with color-coded cards (6 colors), search, grid/list view toggle, and per-note audio playback — saved/starred transcriptions become visual notes.
Choose AI dictation style presets organized by category: Personal, Work, Email, Other. Each style (Casual, Professional, Technical, Creative, etc.) transforms your raw transcription with LLM post-processing.
- General — Shortcut binding, language, microphone, audio feedback, push-to-talk
- Advanced — Autostart, overlay, clipboard handling, model unload timeout, AI post-processing toggle
- Cloud — Provider cards for Groq, Deepgram, and custom self-hosted servers with status indicators
- Post-Processing — LLM provider config, model selection, API keys, custom prompts
- History — Full transcription log with audio playback, copy, star/save, and delete
- Debug — Log level, sound themes, thresholds, recording retention, advanced toggles
- About — App version, language, data directory, credits, and links
📷 Screenshots coming soon — The Tauri v2 app is built and running. Take screenshots with
bun run tauri devinapps/tauri-v2/.
| Feature | SONU | Wispr Flow | Superwhisper | macOS Dictation |
|---|---|---|---|---|
| Fully offline | ✅ | ❌ | ✅ | Partial |
| Open source | ✅ | ❌ | ❌ | ❌ |
| Free forever | ✅ | ❌ ($10/mo) | ❌ ($8/mo) | ✅ |
| Windows + macOS + Linux | ✅ | macOS only | macOS only | macOS only |
| 50+ languages | ✅ | ✅ | ✅ | ✅ |
| Custom dictionary | ✅ | ❌ | ❌ | ❌ |
| Text snippets | ✅ | ❌ | ❌ | ❌ |
| AI text enhancement | ✅ | ✅ | ✅ | ❌ |
| Offline LLM support | 🚧 planned | ❌ | ❌ | ❌ |
| Context-aware dictation | ✅ (Windows) | ❌ | ✅ | ❌ |
| Command Mode (voice rewrite) | ✅ | ❌ | ❌ | ❌ |
| Cloud transcription option | ✅ | ✅ | ✅ | ✅ |
| Custom API endpoint / self-hosted | ✅ | ❌ | ❌ | ❌ |
| Voice notes | ✅ | ❌ | ✅ | ❌ |
| Push-to-talk + toggle | ✅ | ✅ | ✅ | ✅ |
| Auto-type into any app | ✅ | ✅ | ✅ | ✅ |
| Multiple ASR models | ✅ (Parakeet, Whisper & Moonshine local) | ❌ | ✅ | ❌ |
| Themes & customization | ✅ | Limited | Limited | ❌ |
| Platform | Download | Architecture |
|---|---|---|
 |
Download Installer (.exe) | x64, ARM64 |
 |
Download DMG | Intel (x64) + Apple Silicon (ARM64) |
 |
Download AppImage / .deb / .rpm | x64 |
Windows
- Download the
.exeinstaller from Releases - Run the installer and follow the prompts
- Launch SONU from the Start Menu or system tray
- Press your hotkey (default:
Alton Windows,Option+Spaceon macOS,Ctrl+Spaceon Linux) and start speaking
macOS
- Download the
.dmgfrom Releases - Open the DMG and drag SONU to Applications
- Grant Accessibility permissions when prompted
- Press your hotkey and start speaking
Linux
- Download
.AppImage(portable) or.deb(Debian/Ubuntu) from Releases - For AppImage:
chmod +x SONU-*.AppImage && ./SONU-*.AppImage - For .deb:
sudo dpkg -i sonu_*.deb - Press your hotkey and start speaking
SONU supports multiple speech recognition engines and models:
| Model | Download Size | Speed | Accuracy | Best For |
|---|---|---|---|---|
| Moonshine Base | 58 MB | ⚡⚡⚡⚡⚡ | ★★★☆☆ | Ultra-light English dictation |
| Whisper Tiny | 75 MB | ⚡⚡⚡⚡⚡ | ★★☆☆☆ | Quick tests, very old hardware |
| Whisper Base | 142 MB | ⚡⚡⚡⚡ | ★★★☆☆ | Lightweight multilingual dictation |
| Whisper Small | 487 MB | ⚡⚡⚡⚡ | ★★★☆☆ | Balanced multilingual dictation |
| Whisper Medium | 492 MB | ⚡⚡⚡ | ★★★★☆ | High-accuracy work (quantized) |
| Whisper Large | 1.1 GB | ⚡ | ★★★★★ | Maximum accuracy (quantized) |
| Whisper Turbo | 1.6 GB | ⚡⚡ | ★★★★☆ | Large-v3 speed/accuracy balance |
| Parakeet V2 ⭐ | 473 MB | ⚡⚡⚡⚡ | ★★★★★ | English — best speed/accuracy ratio |
| Parakeet V3 ⭐ | 478 MB | ⚡⚡⚡⚡ | ★★★★☆ | Multilingual — default pick |
⭐ Recommended by SONU (Parakeet V3 is the default for new installs; V2 for English-only). Download sizes reflect the quantized builds served by SONU's model catalog. All models run locally on your device — they download automatically on first use.
SONU/
├── apps/
│ └── tauri-v2/ 🦀 Tauri v2 desktop app (Rust + React)
│ ├── src/ React/TypeScript frontend
│ └── src-tauri/ Rust backend (transcription, audio, models)
│
├── docs/ 📚 Documentation & guides
└── .github/ ⚙️ CI, build & release workflows
| Layer | Technology |
|---|---|
| Desktop Framework | Tauri v2 (Rust) |
| Frontend | React 18, TypeScript, TailwindCSS |
| Speech Engine | transcribe-rs (Parakeet TDT · Whisper · Moonshine) |
| AI Enhancement | Cloud providers (OpenAI-compatible, Groq, etc.) |
| Cloud Transcription | Groq, Deepgram, Custom API endpoints |
| Security | OS Keychain, Tauri capability scoping, CSP |
| Testing | Vitest, Playwright, Rust tests, GitHub Actions CI |
- Bun (package manager) — bun.sh
- Rust toolchain — rustup.rs
- Tauri prerequisites — tauri.app/start/prerequisites
# Clone the repository
git clone https://github.com/ai-dev-2024/sonu.git
cd sonu/apps/tauri-v2
# Install dependencies
bun install
# Run in development
bun run tauri dev
# Build for production
bun run tauri buildbun run dev # Start Vite dev server
bun run tauri dev # Start full Tauri dev environment
bun run build # Build frontend
bun run tauri build # Build production binary
bun run test # Run Vitest unit tests
bun run test:e2e # Run Playwright E2E tests
bun run lint # ESLint check
bun run format # Prettier format
bun run typecheck # TypeScript checkSONU can send audio to any transcription endpoint you control. Go to Settings → Cloud, pick the Custom / Self-Hosted provider, and point it at your own OpenAI-compatible API endpoint. Nothing is routed through us.
SONU is designed with security-first principles:
- 🔒 No telemetry — Zero data collection, no analytics, no phone-home
- 🔐 OS Keychain — API keys stored in your OS's secure credential store
- 🧱 Process isolation — Tauri renders in a separate webview process from the Rust backend
- 🛡️ CSP Headers — Content Security Policy limits what the webview can load
- 📁 Path handling — Recording file names are validated before being resolved on disk
Egress note: SONU makes no telemetry or analytics calls. The only outbound requests are model downloads (from the URLs in
resources/models.json) and, if you enable them, update checks and cloud transcription / post-processing against the provider you configure.
- Offline voice-to-text (Parakeet, Whisper & Moonshine)
- AI text enhancement (cloud LLMs; local models planned)
- Context-aware dictation (adapts to the active app)
- Command Mode (voice-rewrite selected text)
- Cloud transcription (Groq, Deepgram, custom endpoint)
- Custom dictionary & text snippets
- Voice notes with search & playback
- Multi-theme support (dark, light, system) with six accent presets
- 50+ language support with auto-detection (cloud Whisper; offline Parakeet covers English + European languages)
- Cross-platform support (Windows, macOS, Linux)
- Custom model fine-tuning
- Plugin / extension system
- Team collaboration features
- Mobile companion app
- Browser extension
- Cloud sync (optional, encrypted)
We welcome contributions! Whether it's bug fixes, features, translations, or docs:
- Fork the repository
- Create a feature branch:
git checkout -b feature/amazing-feature - Make your changes and add tests
- Run checks:
bun run lint && bun run test && bun run typecheck - Commit:
git commit -m "feat: add amazing feature" - Push and open a Pull Request
See AGENTS.md for development guidelines and coding conventions.
| Document | Description |
|---|---|
| AGENTS.md | AI assistant guidelines & build commands |
| CHANGELOG.md | Version history & release notes |
| ARCHITECTURE.md | Technical architecture overview |
| INSTALL.md | Installation guide |
| docs/AI_FEATURES.md | AI post-processing & context-aware dictation |
| docs/BRAND_GUIDELINES.md | Brand & logo usage |
| AUDIT.md | Codebase audit & findings |
| IMPROVEMENT_PLAN.md | Prioritised remediation plan |
| CONTRIBUTING.md | Contribution guidelines |
MIT License — free for personal and commercial use.
- Handy — SONU's foundation; a superb open-source speech-to-text app by CJ Pais
- whisper.cpp — Fast C++ Whisper inference
- Tauri — Secure, lightweight desktop framework
- NVIDIA Parakeet — High-accuracy English ASR
Fork notice: SONU is an independent fork of Handy. The Handy name, logo, and brand assets are not open-source — SONU uses its own branding and does not imply endorsement or affiliation.
Made with ❤️ for people who think faster than they type.
SONU is not affiliated with OpenAI. Whisper is a trademark of OpenAI.