Install

Requirements

macOS 12 or newer — Apple Silicon (M1 or later)
VS Code 1.80+ (Cursor and VSCodium also work)
Claude Code installed and accessible as claude in your PATH
A microphone (built-in is fine; Bluetooth headsets work)

From the Marketplace

Open VS Code → Extensions panel (⇧⌘X).
Search for Verbative and click Install.
A new Verbativeactivity-bar item appears with the Verbative “V” icon. Click it.
The first time you start the listener, macOS will prompt for microphone access. Grant it under System Settings → Privacy & Security → Microphone.

First-run setup

The first time you open the panel, Verbative runs a one-time setup that checks prerequisites, hooks into Claude Code, sets up the voice shortcut, and downloads the on-device engines and models. It tracks each step and then gets out of the way — nothing runs until you start the listener.

Set up Verbative

One-time install of the on-device voice daemon, whisper.cpp and Kokoro TTS. Nothing runs until you start it.

Files & settings

Check prerequisites

Download the on-device engines~150 MB

Hook Verbative into Claude Code

Set up the voice shortcut

Listening & transcription

Download command model~466 MB

Download dictation model~1.6 GB

Voice output

Download voice model~315 MB

A replica of the panel's setup card — the one-time install runs step by step, then gets out of the way.

No signup required

The first time the extension talks to our server it auto-registers a free, device-bound license for you — no email, no paste, no dashboard hop. You can start using voice commands immediately. The license pill in the side panel shows your remaining daily quota.

When you're ready to upgrade to Advanced (or want to use the same account on a second machine), click Sign in / Upgrade in the panel, enter your email, and paste the 8-digit code we email back. You stay on the free tier until you explicitly upgrade.

Sanity check

Click Start listener in the Verbative panel and wait for the spoken cue: “Ready. Say listen to begin.” Then say “listen”. You should hear “listening” and the equalizer should come alive and react to your voice. Speak your prompt, then say “stop listening” to submit. If nothing happens, head to troubleshooting.