Voice for GitHub Copilot CLI

Talk to your Copilot. Out loud, hands-free.

Vox is a Copilot CLI extension that gives any session a voice. Run /vox, a reactive orb opens in its own window, and you just speak — it listens, the agent thinks, and the reply is read back to you.

Pure JavaScript · one slash-command · sibling to Engram & Anya

Windows · PowerShell
> irm https://raw.githubusercontent.com/aasis21/vox/main/install.ps1 | iex
then, in any session  /vox  — tap the orb and talk
The Vox panel listening: a glowing green orb, a live caption of the spoken turn, and the active session in the top dropdown.
MIT licensed Pure JavaScript, no build Windows · macOS · Linux Built on Web Speech No telemetry One /vox away
01 · What it does

A voice canvas for any session.

Run /vox and a single, focused UI comes up with a reactive listening orb. Spoken turns route to the active session; replies are read back. No editor, no tab-juggling — just talk.

  1. 01

    Voice in.

    Speak your turn. Your mic streams straight into the active Copilot session — no typing, no copy-paste.

  2. 02

    Voice out.

    Hear every reply. Vox synthesizes the agent's answer and reads it aloud, so your eyes stay on the work.

  3. 03

    A reactive orb.

    One living orb tells you everything — at rest, hearing you, thinking, or speaking, each with its own color and motion.

  4. 04

    Hands-free flow.

    Grant the mic once and Vox opens straight into listening. Speak, pause, and it sends on its own.

  5. 05

    Interrupt to talk.

    Barge in any time. Tap the orb, press Esc, or hit Interrupt to cut the agent off and drop back into listening.

  6. 06

    Speaks typed replies too.

    Prefer to type? Do it in the CLI — Vox still reads the assistant's reply aloud in the panel.

  7. 07

    Transcript on demand.

    Open the 📜 panel to read the full back-and-forth, scroll the history, or clear it — without losing your place.

  8. 08

    Follows your sessions.

    Many sessions can be live at once. The dropdown lists each one, and running /vox elsewhere auto-switches the window to it.

  9. 09

    Its own window, any OS.

    Opens as a standalone, chrome-less app window — not a browser tab. Pure JS, one-line install on Windows, macOS & Linux.

02 · See it in action

The orb speaks state.

Every moment of the conversation has a look. You always know whether Vox is hearing you, thinking, or talking — at a glance.

At rest

Periwinkle and slowly swirling — tap to begin.

Listening

Green and pulsing with your mic level.

Thinking

Amber while the agent works on it.

Speaking

Blue while the reply is read aloud.

03 · Install

One line. That's it.

Requires Node.js and git on PATH. The bootstrap clones the repo and copies the extension into ~/.copilot/extensions/vox, where Copilot CLI auto-discovers it.

pwsh
> irm https://raw.githubusercontent.com/aasis21/vox/main/install.ps1 | iex

› cloning aasis21/vox → ~/vox
› copying extension → ~/.copilot/extensions/vox  ✓
› node check → v20+  ✓

done. start a Copilot session and run /vox.
04 · Under the hood

Simple machinery. No cloud.

A slash-command, a tiny local server, and a real browser window for the Web Speech APIs. Everything runs on your machine.

Copilot session /vox registers the session · routes your spoken & typed turns
Vox server port 4321 · registry.json · /listen reply channel
App window Chromium app mode · Web Speech in/out · own mic profile

The panel opens as a standalone Chromium app window (Chrome, then Edge) so the Web Speech APIs work where Electron and native webviews can't.
Many sessions can be live at once; a monotonic focus token switches the window to whichever session ran /vox last. Override the browser with VOX_BROWSER.

05 · Commands

Three you'll ever need.

CommandWhat it does
/voxStart voice mode and make this session the active voice target. Opens the panel as its own window (falls back to localhost:4321).
/vox-stopStop Vox for this session and release its voice server.
/vox-whoList live Vox sessions and show which one is active.

Stop typing. Start talking.