Claude Code voice mode in VS Code: from dictation to fully hands-free

You can talk to the Claude Code CLI today — in the terminal and inside VS Code. The real question is which kind of talking you're after: dictation that saves you some typing, or a loop that keeps your hands off the keyboard entirely. Here's what the built-in voice mode does, where it stops, and how to go the rest of the way.

The built-in voice mode: push-to-talk dictation

The Claude Code CLI ships with voice dictation. Type /voice to turn it on, then hold Space, speak, and let go — the transcript lands in the prompt and you press Enter to send it (an autoSubmit setting sends on release, as long as the transcript is at least three words long). There's also a tap mode: with an empty prompt, tap once to start recording, speak, tap again to send — the same three-word rule applies to the auto-send. Recording stops on its own after 15 seconds of silence or two minutes in total.

It works in the terminal and in Anthropic's own VS Code extension, and it's genuinely useful for anyone who thinks faster than they type. A few things to know before you rely on it, straight from the docs as of this writing:

• It needs a Claude.ai sign-in. If you run the Claude Code CLI with an API key or through Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry, /voice isn't available.
• Your audio goes to Anthropic's servers for transcription — it isn't processed on your machine.
• It needs a local microphone, so it doesn't work in the CLI's cloud sessions, over SSH, or in VS Code Remote sessions such as Dev Containers and Codespaces.
• It understands 20 languages, picked up from the CLI's language setting, and falls back to English otherwise.
• It doesn't count against your usage limits.

Where dictation stops

Dictation is input. Everything else about working with an agent is still a keyboard job:

• Nothing is spoken back. You read the screen to find out what Claude did, whether it's finished, or whether it's stuck.
• There's no wake word. A key has to be held or tapped, so your hands stay near the keyboard by design.
• Permission prompts are still typed. Auto mode clears the routine ones, but the prompts that remain — the ones worth seeing — still need a key. The shortcut many people take is bypass mode, which also removes the seeing.
• One prompt in, one session. No voice path for "are you still working?", for aborting a run, or for switching between several agents.
• It's cloud transcription. Fine for most people; a problem if your audio isn't allowed to leave the machine, or if you're one of the API-key users it doesn't cover at all.

If all you want is to type less, the built-in mode is enough — it's already installed. If you want to work away from the keyboard, you need the rest of the loop.

Going fully hands-free in VS Code

Verbative is a VS Code extension that puts a voice layer around the Claude Code CLI — not a replacement for it, a loop on top of it. Put on headphones and it looks like this:

1. Say "listen". The wake word is detected on your machine by a tiny whisper.cpp model, so nothing is recording until you ask.
2. Dictate your prompt. It's transcribed on your machine with whisper.cpp — in roughly 99 languages, translated to English before it reaches Claude — fast enough that the text is ready about when you stop speaking.
3. Say "stop listening". The Claude Code CLI runs the prompt against your codebase.
4. Hear the result. Verbative speaks a short summary back in a natural neural voice — and reads out every permission prompt, so you answer "approve", "deny", or "explain" out loud. You keep the speed of hands-free without going blind.

Along the way you can say "update" for a spoken status check while Claude is working, "stop claude" to abort a run, and switch between a whole team of agents by voice. The keyboard becomes optional, not required.

Two details matter for the comparison above:

• Local by default. Wake-word detection and the full prompt are transcribed on-device, so no audio leaves your machine and our servers never see your voice, your code, or your prompts. Prefer the Claude Code CLI's own cloud dictation? Switch Verbative to Cloud mode and it drives that for you instead, keeping the wake word, the spoken replies, and the voice-answered permissions around it.
• Nothing extra to sign up for. It runs on your existing Claude Pro or Max subscription through the Claude Code CLI — no API key, no SDK, and the free tier activates itself. macOS and Windows.

Which one should you use?

• You just want to type less: use /voice. It's already there, and it's good.
• You want to work from the couch, keep control of every permission, or keep audio on your machine: use Verbative.
• You like the built-in transcription but want the loop: use both — Verbative in Cloud mode.

Set it up in a minute

1. Install Verbative from the VS Code Marketplace.
2. Click the Verbative "V" in the activity bar. A small on-device runtime installs by itself.
3. The free tier activates itself — no sign-up, no API key.
4. Click Start listener. The first time, it tells you what it is about to download — the speech engines and models, about 1.3 GB — and keeps them on your machine. Then say "listen".

The voice docs cover the commands, the Local and Cloud modes, and the whisper models you can pick from.

Verbative is voice-first, on-device tooling for AI coding agents. See how the voice loop works →

Frequently asked

Does Claude Code have a voice mode in VS Code?+

Yes, push-to-talk dictation. Enable it with /voice, hold Space or use tap mode, and the transcript lands in the prompt. It works in the terminal and in Anthropic's VS Code extension, needs a Claude.ai sign-in, and streams your audio to Anthropic's servers. It does not speak replies, listen for a wake word, or take permission answers by voice.

Does Claude Code's dictation work with an API key?+

No. The built-in dictation requires a Claude.ai account as the active sign-in. With an API key, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry configured, the /voice command is not available.

Can Claude Code talk back or read its responses aloud?+

Not built in. Verbative adds spoken output around the Claude Code CLI in VS Code: a short summary when a task finishes, permission prompts read aloud so you can say approve, deny or explain, and a spoken status check when you say update. The voice is generated on your machine.

Is my voice processed locally?+

With the built-in dictation, no: audio is sent to Anthropic for transcription. With Verbative's default Local mode, wake-word detection and the full prompt are transcribed on your machine with whisper.cpp, so no audio leaves it. Cloud mode uses the CLI's own dictation instead if you prefer it.