Whisper Flow 🎙️

A lightweight, local AI voice transcription tool built with Tauri v2, React, and Rust. Whisper Flow runs OpenAI's Whisper models locally on your machine, ensuring privacy and speed without sending audio data to the cloud.

📥 Download

Note: Whisper Flow is currently available for macOS (Silicon/Intel) only.

Download the latest version from GitHub Releases

✨ Features

• 🔒 Local & Private: All transcription happens on-device using quantized Whisper models (ggml).
• 👻 Native Floating Hint: A beautiful, completely transparent floating capsule indicates recording status without blocking your workflow.
  • Built with native macOS NSWindow APIs for true transparency and click-through capability.
• ⚡️ Global Shortcuts: Toggle recording instantly from anywhere (Default: Shift + Command + A).
• 📂 Drag & Drop: Drag audio/video files directly into the app to transcribe them.
• 📋 Auto-Copy: Transcribed text is automatically copied to your clipboard.
• 🎯 System Integration:
  • Native microphone access.
  • Accessibility API integration for global input monitoring.
  • macOS "Ghost Window" mode (ignores mouse events, fully transparent background).

🛠️ Tech Stack

• Frontend: React, TypeScript, Vite, Modular CSS
• Backend: Rust (Tauri v2), objc2 (for macOS native window management), cpal (Audio), whisper.cpp (Model inference)
• State Management: React Hooks + Tauri Event System

Getting Started

Prerequisites

• Node.js (v18+)
• Rust (latest stable)
• macOS (tested on Sonoma/Sequoia)

Installation

1. Clone the repository

   git clone https://github.com/yourusername/whisper-flow.git
   cd whisper-flow

2. Install dependencies

   npm install

3. Setup Binaries & Models

   • Models: The app will automatically download the Whisper model on first run.
   • Binaries: You must manually place the required binaries in src-tauri/binaries/:
     • ffmpeg: Download a standalone FFmpeg binary (aarch64 for Apple Silicon, x64 for Intel).
     • whisper-cli: Build or download whisper-cpp CLI.
     • Naming: Ensure files are named specifically for your architecture, e.g., ffmpeg-aarch64-apple-darwin and whisper-cli-aarch64-apple-darwin.

4. Run in Development Mode

   npm run tauri dev

5. Build for Production

   npm run tauri build

🧩 Permissions

On the first launch, the app will request the following permissions:

1. Microphone: To record audio.
2. Accessibility: To listen for global shortcuts (e.g., Shift+Cmd+A) even when the app is in the background.

🏗️ Architecture Highlights

• Multi-Window System: Uses a dedicated lightweight hint.html entry point for the floating window to minimize resource usage.
• Zero-Style-Leakage: The floating window uses a separate CSS stack (hint.css) and Native DOM injection to prevent main window styles from affecting the transparent background.
• Rust-Native Transparency: Leverages objc2 and Cocoa APIs to manipulate the underlying NSWindow, ensuring a "glass-like" effect that standard Webviews cannot achieve alone.

📝 License

MIT License