Claude Code voice: when /voice is enough, and when to use system dictation
Claude Code has voice built in: run /voice, hold Space and speak, and the words appear in its prompt. According to Anthropic’s documentation, the recording is sent to Anthropic’s servers for transcription, and it needs a Claude.ai account and a local session. Dalm is a different tool: a Mac app where you hold one key in any app — the terminal running Claude Code, your editor, a pull request in the browser, a chat — and the words are typed where your cursor is. The speech is recognised on your Mac and no audio is sent anywhere; the text of every dictation is synced to our servers automatically. There is a Free plan.
Claude Code voice mode: what /voice does
Claude Code’s voice mode is a command: type /voice and dictation stays on from one session to the next. By default it is push-to-talk — hold Space, speak, let go — and with /voice tap one tap starts recording and the next one sends the prompt. The words show up in the prompt while you talk and settle when you stop; in hold mode the prompt then waits for Enter unless auto-submit is on. Recognition is tuned for programming words, and the project name and git branch are passed in as hints. Dictation follows Claude Code’s language setting: English, unless you pick one of 20 supported languages. According to Anthropic’s documentation, the recorded audio is streamed to Anthropic’s servers for transcription, it needs a Claude.ai account and a microphone on the machine you are at, and it does not count toward your usage limits. Besides the terminal, the documentation describes it in agent view and in the VS Code extension.
When the built-in /voice is enough
If everything you say is meant for Claude Code’s prompt, start with /voice. There is nothing to install, it uses none of your usage, it already knows the name of your project, and it can send the prompt for you. It fits when you sign in to Claude Code with a Claude.ai account, work on your own machine rather than over SSH or in a cloud session, speak one of its 20 languages, and are comfortable with the audio going to Anthropic to be transcribed. When all of that holds, a separate dictation app adds a second key and not much else for this one job.
Claude Code voice input in the terminal and outside it
Dictation that works across the whole Mac earns its place in three cases.
- Claude Code signs in some other way. Anthropic’s documentation says voice dictation is not available with an API key, Amazon Bedrock, Google Cloud’s Agent Platform or Microsoft Foundry. Dalm treats Claude Code like any other app: it types into the window in front, so the way Claude Code is signed in does not matter to it.
- The words are not only for the prompt. A pull request description, a ticket, a commit message in your editor, a reply to a colleague: one key covers each, because the text goes wherever the cursor is. /voice is built for Claude Code’s own input.
- You would rather the audio stayed on your Mac. Dalm recognises speech on the Mac and sends no audio anywhere; what it does send is text — every dictation is synced to our servers automatically, with no off switch. There is no language to set first either: it is detected as you speak, within the range picked at setup, 25 European languages or more than a hundred.
By default the two listen on different keys — Space in Claude Code, the right ⌘ in Dalm.
Dictating into Claude Code with Dalm
Click into Claude Code’s prompt, hold the right ⌘ key, say what you want done, and release: the words arrive as keystrokes, and whatever you had copied stays on the clipboard. Nothing presses Enter, so the prompt sits there until you have read it and sent it yourself. For a longer instruction, tap the key once, talk the task through, and tap again. Add the names that keep coming out wrong — a library, an internal service, a teammate — to the dictionary along with how they come out, and they are corrected from then on. Typing was checked by hand in Telegram, Notion, TextEdit and Safari; the terminal and code editors receive the same keystrokes but were not part of those checks. Whatever happens in the window, each dictation is also kept in the Dictations list, ready to copy. Setup, step by step, is on voice to text on a Mac.
Audio and text: what each one sends, and where
| Claude Code /voice | Dalm | |
|---|---|---|
| Your voice | Streamed to Anthropic’s servers for transcription, per Anthropic’s documentation | Recognised on your Mac; no audio is sent anywhere, and it is deleted once the text exists |
| The words | Inserted into Claude Code’s prompt | Typed into the field you are in, and synced to our servers automatically |
| A language model | The prompt goes to Claude when you send it | Your dictation is not sent to a language model; the prompt goes to Claude when you send it, as with anything typed |
Either way, once you press Enter the prompt goes to Claude: what differs is where your voice is turned into text. In Dalm nobody else sees your dictations, even in a team. Everything that leaves the Mac, piece by piece, is on privacy; the trade-off between recognising speech on the device and in the cloud in general is on its own page.
Trying it: Free, Pro or Team
Dictation comes with the Free plan, and Free is a plan rather than a trial. Dalm needs a Mac with Apple Silicon · macOS 15.0 or later and an account; while it has not been released, the way in is the waitlist. What the paid plans add is on the pricing page. For a team working on one codebase, the dictionary — service names, product terms, the colleague whose surname never comes out right — is shared by everyone, while each person’s dictations stay their own: working as a team.
Questions
- Does Claude Code have voice input?
- Yes. Run /voice in Claude Code, hold Space and speak, and the words appear in the prompt. Anthropic's documentation describes it in the terminal, in agent view and in the VS Code extension; it needs a Claude.ai account and a microphone on the machine you are at, so it is not there in SSH or cloud sessions.
- How do I use voice mode in Claude Code?
- Type /voice to switch it on; on a Mac, the first run asks for microphone access for your terminal. Hold Space while you speak and let go to finish, or switch to /voice tap and use one tap to start and one to send. The dictation language comes from /config: English unless you choose one of the 20 supported languages. Esc or Ctrl+C throws a recording away.
- Can I dictate to Claude Code on a Mac without /voice?
- Yes, with dictation that works across the system. Dalm types your words as keystrokes into the window in front, the terminal running Claude Code included, and does not press Enter, so the prompt waits for you. The terminal and code editors were not among the apps tried by hand — those were Telegram, Notion, TextEdit and Safari. Dalm needs a Mac with Apple Silicon · macOS 15.0 or later and an account, and while it has not been released, the way in is the waitlist.
- Where does Claude Code speech to text happen?
- On Anthropic's servers: according to Anthropic's documentation, /voice streams the recorded audio there for transcription, and the audio is not processed locally. With Dalm, the speech is recognised on your Mac and no audio is sent anywhere; the text of each dictation is synced to our servers automatically, and the prompt reaches Claude when you send it, however it was written.
- Is there Claude Code dictation if I use an API key or Bedrock?
- Not through /voice: Anthropic's documentation says voice dictation needs a Claude.ai account and is not available with an API key, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. Dalm works with Claude Code the way it works with any app — it types into the terminal window — so how Claude Code is signed in makes no difference to it.
- Can I talk to Claude Code instead of typing?
- Yes, in the sense of speaking your prompts: /voice and Dalm both turn what you say into the text of a prompt, and Claude Code works from that text. With /voice the words land in Claude Code's prompt; with Dalm the same key also covers your editor, the browser and chat. Both have a tap mode for a long prompt — with /voice the second tap also sends it, with Dalm it only types it.
- Does voice dictation in Claude Code cost anything?
- According to Anthropic's documentation, /voice transcription uses no Claude messages or tokens and does not count toward the limits in /usage. Dalm dictation is part of the Free plan, which is a plan rather than a trial; what the paid plans add is on the pricing page.