Vowrite 🎙️

AI Voice Keyboard for macOS — Speak naturally, get polished text inserted at your cursor.

🇨🇳 中文文档

Vowrite is a lightweight macOS menu bar app that turns your voice into clean, polished text — inserted right where your cursor is. Powered by OpenAI Whisper for transcription and GPT for text polishing.

No more typing. Just speak.

---

✨ Features

• 🎤 Voice-to-Text — Press a hotkey, speak, get text
• ✨ AI Polish — Automatically removes filler words, fixes grammar, adds punctuation
• 🌍 Multilingual — Chinese, English, and mixed-language support
• 📋 Smart Injection — Text appears directly at your cursor position
• 🎯 Works Everywhere — Tested in native apps, browsers, Discord, VS Code, and more
• 🎨 Floating Overlay — Compact recording bar with smooth waveform animation
• ⌨️ Customizable Hotkey — Default: ⌥ Space (Option + Space)
• ⎋ ESC to Cancel — Press Escape to instantly cancel recording
• 📊 History — Browse and search past dictations
• 🔌 Multi-Provider — OpenAI, OpenRouter, Groq, Together AI, DeepSeek, or bring your own

🚀 Quick Start

Download

Download the latest release from GitHub Releases.

Build from Source

cd VowriteApp
./build.sh

Or manually:

cd VowriteApp
swift build -c release
cp .build/arm64-apple-macosx/release/Vowrite Vowrite.app/Contents/MacOS/Vowrite
codesign -fs - --deep --entitlements Resources/Vowrite.entitlements Vowrite.app
open Vowrite.app

Setup

1. Launch Vowrite — it appears in the menu bar as a 🎤 icon
2. Open Settings → enter your API key (Get one from OpenAI)
3. Grant Microphone and Accessibility permissions when prompted
4. Press ⌥ Space to start recording, press again to stop
5. Text is automatically inserted at your cursor ✨

🔧 How It Works

🎤 Record → 📝 Transcribe → ✨ Polish → 📋 Insert

1. Record — Audio captured as AAC via AVAudioEngine
2. Transcribe — Sent to Whisper API for speech-to-text
3. Polish — GPT cleans up filler words, grammar, and punctuation
4. Insert — Text injected at your cursor via clipboard paste or Unicode typing

Text Injection Methods

Method	Speed	Requires
Clipboard Paste (default)	⚡ Instant	Accessibility permission
Unicode Typing (fallback)	Fast	Nothing — works everywhere

Vowrite automatically detects permissions and picks the best method.

🔌 Supported Providers

Provider	STT Model	Polish Model
OpenAI	whisper-1	gpt-4o-mini
OpenRouter	whisper-large-v3	gpt-4o-mini
Groq	whisper-large-v3-turbo	llama-3.1-8b-instant
Together AI	whisper-large-v3	Llama-3.1-8B-Instruct-Turbo
DeepSeek	whisper-1	deepseek-chat
Custom	configurable	configurable

📁 Project Structure

VowriteApp/
├── App/                        # App lifecycle & state
├── Core/
│   ├── Audio/                  # Microphone recording
│   ├── STT/                    # Speech-to-text (Whisper)
│   ├── AI/                     # Text polishing (GPT)
│   ├── TextInjection/          # Cursor text injection
│   ├── Hotkey/                 # Global hotkey management
│   └── Keychain/               # Secure API key storage
├── Views/                      # SwiftUI views
├── Models/                     # SwiftData models
├── Resources/                  # Info.plist, entitlements
└── build.sh                    # Build script

📋 Requirements

• macOS 14.0 (Sonoma) or later
• API key from a supported provider (OpenAI recommended)
• Microphone permission
• Accessibility permission (recommended, not required)

🗺️ Roadmap

• v0.1 — Core voice dictation
• v0.2 — Release packaging & error handling
• v0.3 — App icon & branding
• v0.4 — Custom prompts, multiple output modes
• v0.4 — Real-time streaming, local Whisper
• v1.0 — Code signing, notarization, auto-update

See full roadmap for details.

📄 License

MIT License — see LICENSE for details.

🤝 Contributing

Contributions welcome! Please open an issue first to discuss what you'd like to change.

📝 Changelog

v0.3 — App Icon

• Official app icon (waveform ring + text cursor design)
• Icon generation automation script
• Build pipeline integration

v0.2 — Release Ready

• Release build optimization (debug logs disabled in production)
• User-friendly error messages
• Automated DMG packaging

v0.1 — Initial Release

• Voice dictation with Whisper STT + GPT polish
• Menu bar app with floating recording overlay
• Customizable hotkey (default ⌥Space)
• Multi-provider support
• Dual text injection (clipboard paste + Unicode typing fallback)
• Dictation history with SwiftData
• Microphone selection & launch at login