Vox

Product Introduction

1. Definition: Vox is a voice-first, local-first extension for GitHub Copilot CLI and the GitHub Copilot desktop application. Technically, it is a JavaScript-based CLI extension that launches a standalone Chromium application window to provide a bidirectional voice interface for AI pair programming.
2. Core Value Proposition: Vox exists to enable a truly hands-free, eyes-free workflow with GitHub Copilot. Its primary value is transforming the AI coding assistant interaction from a text-based, screen-locked experience into a reactive, auditory conversation, allowing developers to maintain focus on their primary task—whether that's writing code, debugging, or architectural thinking—without constant context switching.

Main Features

1. Reactive Voice Canvas: The core UI is a single, stateful orb that opens in its own chromeless application window. It visually and audibly communicates its state (At Rest, Listening, Thinking, Speaking) through color and motion, providing immediate, glanceable feedback without distracting overlays or complex interfaces.
2. Full-Duplex Voice Interaction: Vox utilizes the Web Speech API for both Speech-to-Text (STT) and Text-to-Speech (TTS). It streams microphone input directly to the active Copilot session and synthesizes the agent's textual replies into spoken audio. The system features automatic voice activity detection (VAD) to send turns upon a pause and supports real-time interruption (via Esc key or orb tap) to barge in.
3. Multi-Session Awareness & Management: The extension integrates deeply with the Copilot CLI session layer. It maintains a registry of active sessions, allows users to switch the active voice target via a dropdown, and automatically focuses the Vox window on the session where /vox was most recently invoked. This enables seamless context switching across multiple terminal sessions or projects.
4. Local-First, Privacy-Centric Architecture: All processing occurs on the user's machine. Voice recognition and synthesis are handled client-side via the browser's Web Speech API, and communication flows between the local Vox server (port 4321) and the Copilot CLI session. The project explicitly states it contains no telemetry, ensuring all conversation data remains private.
5. Integrated Transcript Panel: A slide-in panel (📜) provides a real-time, scrollable transcript of the entire voice conversation. This allows for reviewing past interactions, copying specific responses, or clearing history without disrupting the current auditory flow, bridging the gap between voice convenience and textual reference.

Problems Solved

1. Pain Point: Context Switching and Flow State Disruption. The constant need to look away from code, type a prompt, wait, and read a response breaks a developer's deep concentration. Vox eliminates this by making the interaction auditory and parallel to the primary visual task.
2. Target Audience: The primary personas are Productive Developers and Engineers (especially those using VS Code or terminals heavily), Accessibility-Focused Users who benefit from voice interfaces, and DevOps/SRE Professionals who might need hands-free assistance while managing systems or consoles.
3. Use Cases: Exploratory Coding & Debugging: Verbally describing a bug or a desired function while examining code. Documentation & Learning: Asking Copilot to explain a concept or library aloud while following along in a browser or book. Accessibility: Enabling developers with repetitive strain injury (RSI) or visual preferences to interact with Copilot efficiently. Multi-Monitor Workflows: Keeping Copilot interactions on a secondary screen or auditory channel while primary screen real estate is dedicated to IDEs, browsers, or terminals.

Unique Advantages

1. Differentiation: Unlike generic voice-typing tools or assistants, Vox is a dedicated, tightly integrated extension for GitHub Copilot. It is not a cloud-based service (like some AI voice assistants) but a local tool. Compared to attempting to use system-wide dictation with Copilot, Vox provides a purpose-built, state-aware interface with direct session routing and reply synthesis.
2. Key Innovation: Its architecture decouples the voice interface from the editor/terminal by using a standalone Chromium app window. This cleverly bypasses the limitations of Electron/webview environments where Web Speech APIs are often restricted or unavailable, ensuring robust cross-platform (Windows, macOS, Linux) voice functionality. The "monotonic focus token" system for multi-session management is also a novel solution for a CLI extension environment.

Frequently Asked Questions (FAQ)

1. Is Vox for GitHub Copilot secure and private? Yes, Vox is a local-first, open-source (MIT licensed) extension. All voice processing uses your browser's local Web Speech API; no audio or conversation data is sent to any cloud service other than what GitHub Copilot itself requires for its core functionality. The extension adds no telemetry.
2. How does Vox work with multiple GitHub Copilot CLI sessions? Vox includes a session registry. When you run /vox in a terminal session, that session becomes the active voice target. The Vox window's dropdown lists all live sessions, and the window will automatically refocus to the session that last invoked /vox, enabling seamless voice control across different projects or contexts.
3. Can I use Vox with the GitHub Copilot app in VS Code? Yes. Vox can be added directly from within the GitHub Copilot app's panel. Once installed, it operates within the same Copilot sidebar panel, providing the voice interface directly inside VS Code without needing a separate /vox command or standalone window.
4. What are the system requirements for installing Vox? Vox requires Node.js (v20+) and git to be installed and available on your system's PATH. The one-line installer (PowerShell or bash) will clone the repository and copy the extension to ~/.copilot/extensions/vox. A compatible Chromium-based browser (Chrome or Edge) is required for the app window.
5. Can I type instead of speak when using Vox? Yes. You can continue to type your prompts directly into the GitHub Copilot CLI or chat interface as usual. Vox will still detect the agent's textual reply and read it aloud to you through its TTS engine, maintaining the voice-out benefit even for typed input.