Skip to content

Developer guide · Checked

Voice dictation for Claude Code.

Claude Code has dictation built in: /voice. Here’s how to turn it on, the setups where it isn’t available, and how to talk to Claude Code in those too, with an app that types into your terminal.

Built in

Turn on /voice.

It needs a Claude.ai sign-in and a microphone on the machine running Claude Code.

  1. Sign in with Claude.ai.

    Run /login if you aren’t. If an API key or a cloud provider is set up, it takes precedence: remove it and restart Claude Code.

  2. Run /voice.

    Claude Code checks your microphone. On a Mac, the first time triggers the permission prompt for your terminal app. The setting stays on across sessions.

  3. Hold Space, talk, release.

    Recording starts after a short warmup while you hold the key, and your words appear in the prompt as you speak. Release to finish, then press Enter to send.

  4. Adjust it if you like.

    /voice tap switches to tap to start, tap to send. You can bind another key in ~/.claude/keybindings.json, and set your dictation language in /config.

Anthropic’s docs add: “Transcription is tuned for coding vocabulary in both modes.” and “Transcription does not consume Claude messages or tokens and does not count toward the limits shown in /usage.” Source: Voice dictation in the Claude Code docs, read October 7, 2026.

Where it stops

When /voice isn’t available.

These come from Anthropic’s own requirements and troubleshooting notes.

You sign in with an API key or a cloud provider

/voice is only available with a Claude.ai sign-in. Anthropic’s docs list an Anthropic API key, Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry as setups where it is not available, and an API key or third-party provider wins over a Claude.ai sign-in. You see “Unknown command: /voice” or “Voice mode requires a Claude.ai account”.

You work on another machine

The microphone has to be on the machine running Claude Code, so /voice does not work in SSH sessions or cloud sessions, nor in VS Code remote sessions (SSH, Dev Containers, and Codespaces). In WSL it needs WSLg.

You speak a language it doesn’t list

Dictation uses Claude Code’s language setting, set in /config, and covers 20 languages. Any other setting falls back to English, and you change the setting to dictate in another language.

You want to dictate everywhere else too

/voice lives in the Claude Code prompt. Commit messages, pull request descriptions, docs and chat are typed elsewhere, where it can’t reach.

An administrator can also turn it off for an organization (“Voice mode is disabled by your organization’s policy”). And because /voice sends audio to Anthropic, some teams may prefer speech that never leaves the laptop. See Anthropic’s data usage page for how it handles your data.

Works anywhere

Dictation that types into your terminal.

System-wide dictation doesn’t talk to Claude Code at all. It turns your speech into text and types it into the window in front, so Claude Code receives keystrokes, as if you had typed. That’s why it doesn’t care how Claude Code is signed in, and why it works over SSH: your terminal sends the keys on to the remote shell. It works the same for other terminal agents, in your editor and in the browser. Developers who direct coding agents every day are the kind of people AI labs pay to build and grade the coding tasks those agents train on, and our other site, AITraining.jobs, explains what that work involves and how to start.

Every computer has one built in. On a Mac, Apple’s Dictation is free; here’s how to set it up. On Windows, voice typing opens with Windows key + H and uses Microsoft’s online speech service; how it compares. Dictation apps such as Wispr Flow and Aqua Voice send your speech to their servers to turn it into text (compare Wispr Flow and Aqua Voice).

Yap is ours: hold Fn on a Mac or Ctrl + Win on Windows, speak and release, and your words are typed at the cursor, with speech recognized on your computer and nothing uploaded. It doesn’t touch your clipboard and never presses Enter, so you read the prompt before it goes to the agent. How it compares with every voice tool for coding.

Working with recordings instead of live speech? Our free, open-source Yap MCP server lets Claude Code, Claude Desktop and other MCP clients transcribe audio files on your computer, with the same speech engine.

Checked on : through its normal typing path, Yap 0.7.38 put a sentence, character for character, into a raw-mode prompt in macOS Terminal, the kind of input a terminal agent like Claude Code reads.

Side by side

/voice or Yap?

Claude Code’s built-in voice dictation compared with Yap
FeatureClaude Code /voiceYap
Where speech becomes textAnthropic’s servers: “Audio is not processed locally.”On your Mac or PC. Audio is not uploaded.
Claude Code sign-inA Claude.ai account only; not with an Anthropic API key, Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft FoundryAny: Claude Code just sees typing. Yap needs its own subscription.
SSH, cloud and remote sessionsNot availableWorks: it types into your local window, which passes the keys on
Where you can dictateThe Claude Code prompt, in the CLI and the VS Code extensionAny app with a text cursor: terminals, editors, browsers, chat
Languages20, one at a time, set in /config25, recognized in whichever you speak, per recording
Coding wordsTuned for coding vocabulary; adds your project and branch names as hintsNo custom vocabulary or coding hints
LengthTap mode stops after 15 seconds of silence or 2 minutesUp to 5 minutes per recording
SendingHold mode waits for Enter unless you turn on autoSubmit; tap mode sends prompts of 3 words or moreNever presses Enter: you read the prompt, then send it
CostIncluded. It doesn’t use Claude messages or tokens.$6.90 USD a month or $69 USD a year, after 3 free days and a 10-day trial with a card

Languages only /voice lists: Hindi, Indonesian, Japanese, Korean, Norwegian, and Turkish. Languages only Yap recognizes: Croatian, Slovenian, Latvian, Lithuanian, Estonian, Finnish, Slovak, Hungarian, Romanian, Bulgarian, and Maltese. On a Mac, Yap’s 25 languages need macOS 14 or later; older versions use compatible local recognition. Claude Code details from Anthropic’s docs, read October 7, 2026. Claude Code is Anthropic’s product; Yap is not affiliated with Anthropic.

Being fair

Which one to use.

Use /voice if

  • You sign in to Claude Code with a Claude.ai account and run it on your own machine.
  • Your prompts are full of library names and identifiers: it is tuned for them, and Yap has no custom vocabulary.
  • You’re fine with your audio going to Anthropic.
  • You only want dictation in Claude Code, and want it free.

Use Yap if

  • You run Claude Code over SSH, or with an Anthropic API key, Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry.
  • Your speech should stay on your computer.
  • You dictate in more than one language, or in one /voice doesn’t list, such as Romanian, Hungarian, or Finnish.
  • You want the same shortcut for prompts, commit messages, docs and chat, on a Mac or a Windows PC.

With Yap

Dictate a prompt in four steps.

  1. Install Yap and finish setup.

    On a Mac, allow Microphone and Accessibility; Accessibility is what lets it type into your terminal. On Windows, allow desktop apps to use the microphone. The first recording downloads the speech model. Setup guide.

  2. Click into the Claude Code prompt.

    In your terminal, the VS Code panel or the browser: wherever you would type. On Windows, use a terminal that isn’t running as administrator: Windows doesn’t let apps type into those, so your words would stay in History.

  3. Hold Fn on a Mac or Ctrl + Win on Windows, speak, release.

    For a long prompt on a Mac, press Space while holding Fn to keep recording hands-free, then press Fn again to finish. Up to 5 minutes per recording.

  4. Read it, then press Enter.

    Your words are typed into the prompt and also saved in Yap’s History on your computer, so nothing is lost if focus moved.

Questions people ask.

Does Claude Code have voice dictation?
Yes. Run /voice in the Claude Code CLI, then hold Space while you speak and release it to finish. Your words appear in the prompt, and you press Enter to send. The VS Code extension supports it too.
Why does Claude Code say “Voice mode requires a Claude.ai account”?
/voice only works while a Claude.ai account is your active sign-in. Anthropic’s docs say it is not available when Claude Code uses an Anthropic API key, Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry, and an API key or a third-party provider takes precedence over a Claude.ai sign-in. In those setups, an app that types into your terminal still lets you dictate.
Does /voice work over SSH?
No. Anthropic’s docs say voice dictation does not work in SSH or cloud sessions, or in VS Code remote sessions (SSH, Dev Containers, and Codespaces). A dictation app on your own computer types into your local terminal window, which sends the keystrokes on to the remote shell like any typing.
Is Claude Code’s voice dictation processed on my computer?
No. Anthropic’s docs say: “Voice dictation streams your recorded audio to Anthropic’s servers for transcription. Audio is not processed locally.” Yap recognizes speech on your Mac or PC instead, and does not upload your audio.
Does Yap work with Codex, Gemini CLI or Cursor?
Yap types into whichever app is in front, so a coding agent in a terminal receives your words the same way Claude Code does, and so do the text fields of editors and browsers. It does not press Enter for you.

From the library

Voice Typing in Google Docs, Word and Every Other App

Google Docs, Word, Outlook, PowerPoint, OneNote, ChatGPT, Claude and Claude Code have their own dictation button, but each one types only inside its own app and usually needs an internet connection or a paid plan. Gmail on a computer, Slack, terminals and Linux desktops have none, so the reliable setup is the built-in button where it is good enough plus one system-wide dictation tool that types into whatever text field has focus.