vox

Vox is a hands-free voice panel extension for GitHub Copilot that lets you interact with the Copilot agent by speaking and hear its replies. It works both with GitHub Copilot CLI and the GitHub Copilot app.

Purpose

Vox's core purpose is to let developers drive GitHub Copilot sessions by voice. Instead of typing manually in the editor, you simply speak, and Vox converts your speech to text, sends it to the active Copilot session, and reads the agent's reply aloud via speech synthesis. It is suited for scenarios where you want to be hands-free, reduce keyboard use, or switch quickly between multiple sessions.

Main capabilities

• Voice input/voice output: Microphone audio is streamed directly to the active Copilot session, and the agent's replies are synthesized and read aloud.
• Reactive orb: A dynamic orb reflects status via color—light purple when idle, green when hearing you speak, amber when thinking, and blue when speaking; the orb also pulses with the microphone volume so you know your voice is being picked up.
• Hands-free flow: If microphone permission has been granted, Vox goes straight into listening mode. A brief pause after you finish speaking sends automatically, or you can click the orb or press the spacebar to send immediately.
• Interrupt speech: While the agent is speaking, clicking the orb, pressing Esc, or clicking Interrupt can interrupt it and return straight to listening mode, without needing to click the microphone again.
• Live captions: Your speech is displayed as streaming temporary text and is progressively confirmed as you speak.
• Read typed replies aloud: Even if you type directly in Copilot CLI, Vox will read the assistant's replies aloud in the panel.
• Transcript panel: Open the 📜 panel to view the full back-and-forth conversation record; it can be closed or cleared at any time.
• Session routing and automatic switching: A centered dropdown shows each active session; multiple sessions can be active at once, and running /vox from another session automatically switches the already-open window to that session.
• Standalone app window: Opens as a borderless desktop-style window via Chrome/Edge's app mode, rather than a browser tab, so the Web Speech API works properly.
• Cross-platform: Pure JavaScript, no build required. One-line installation on Windows, macOS, and Linux.

Usagemethod

1. Prerequisites: Requires GitHub Copilot CLI (command copilot, installable via npm install -g @github/copilot ), plus Node.js and git in PATH.
2. Install: Run the one-line install command in any shell (PowerShell on Windows, bash on macOS/Linux). The command clones the repository to ~/vox(or updates it if it already exists), then copies the extension to ~/.copilot/extensions/vox, where Copilot CLI will automatically discover it.
3. Launch: Start a Copilot session and run /vox. Click the orb or press the spacebar, speak your instruction, and it will be sent after a pause, with the reply read back aloud.
4. Use in the Copilot app: Open Vox from the app's panel (or have Copilot open it). Note that slash commands are only available in CLI, but the panel's functionality is the same.
5. Manual install/development: You can install from a local clone.

Technical details

• The /vox command starts a small local server on port 4321 and registers the session in registry.json . The browser canvas streams microphone audio and plays synthesized replies, routing voice turns to the active session. A persistent /listen channel also streams replies from CLI typing turns to the panel so that they are read aloud too.
• Multiple sessions can be active at once;/vox-who Shows which is the active session. Running from another session /vox sets a monotonic focus token, so an already-open window switches to the most recently requested session without creating duplicate windows.
• /vox Uses an installed Chromium browser (Chrome first, then Edge) to open the panel as a standalone borderless window in app mode. This ensures the browser's Web Speech API works properly, which Electron and native webviews do not support. The window uses a dedicated profile (~/.copilot/vox-app-profile), so it has an independent identity and remembers microphone permissions. It can be overridden via VOX_BROWSER=chrome|edge|brave|chromium or the full path to the executable.

Use cases

• Programmers developing with GitHub Copilot CLI or the Copilot app who want to quickly issue commands and receive feedback by voice.
• Users who need to manage multiple Copilot sessions at the same time and switch quickly between them.
• Scenarios where you want to reduce keyboard use, keep your hands free, or improve interaction efficiency.

Limitations

• GitHub Copilot CLI, Node.js, and git must be installed.
• Slash commands in the Copilot app are unavailable, but the panel has the same functionality.
• It relies on the Chromium browser's Web Speech API, so one of Chrome, Edge, Brave, or Chromium must be installed.
• Microphone permission is required.
• Other feature details or platform support are subject to actual use.