VAKH - Offline AI Dictation

May 18, 2026 ยท View on GitHub

VAKH is a minimalist, high-performance desktop application for real-time voice-to-text dictation. It runs entirely offline using OpenAI's Whisper model, ensuring your voice data never leaves your computer.

VAKH Interface

Features

  • 100% Offline: All processing is done locally. No API keys, no subscriptions, and no data collection.
  • Fast & Responsive: Powered by Rust and Tauri for minimal overhead and low latency.
  • Universal Typing: Works across all applications (Word, Slack, Browser, Terminal, etc.) by simulating keystrokes.
  • Smart Slicing: Automatically flushes text every few seconds or at the end of sentences for a smooth flow.
  • Visual Feedback: Real-time waveform visualization with color-coded states (Blue for listening, Green for speaking, Red for warnings).
  • 5-Minute Sessions: Support for long-form dictation up to 5 minutes per session.

Usage

  1. Launch VAKH: Open the application. A minimalist "orb" UI will appear at the bottom of your screen.
  2. Start Dictating: Double-tap the Left Ctrl or Right Ctrl key to start listening. The UI will turn Blue.
  3. Speak: As you talk, the UI will pulse Green. Your words will appear in the focused text field almost instantly.
  4. Stop/Finalize: Double-tap Ctrl again (or click the stop button) to finalize the transcription.
  5. Hide/Close: Press Esc or click the close button to hide the window.

Installation

Windows

  1. Download the latest vakh_0.1.0_x64_en-US.msi or vakh.exe from the Releases page.
  2. Run the installer.
  3. Launch VAKH from your Start menu.

Build from Source

If you want to build the project yourself:

  1. Install Rust and Node.js.
  2. Clone this repository.
  3. Run npm install.
  4. Run npm run tauri build to generate the production binaries.

Tech Stack

  • Backend: Rust, Tauri 2.0
  • AI Model: Whisper (embedded via whisper-rs)
  • Frontend: Vanilla HTML/JS/CSS
  • Input Simulation: Win32 API (SendInput)

License

This project is licensed under the MIT License - see the LICENSE file for details.

Disclaimer

VAKH uses a small, optimized version of the Whisper model (tiny.en). While very fast, it may occasionally hallucinate or make errors in complex environments.