Voice input in Terminal on Mac.

Why a terminal has no dictation of its own

A terminal is not a text editor with features, it is a place where characters arrive. That is why no terminal ships dictation and why none of them will: the right layer for it is the system, not the app. Anything that can put characters into the focused field can put them into a terminal, which is what system wide dictation does.

What it is actually good for

Be honest about this one, because the demonstration everyone does is the worst case. Saying a short command out loud is slower than typing it and always will be. The gains are elsewhere, and they are large:

• Commit messages with the reasoning in them instead of four words.
• Prompts to a coding agent running in the terminal, where the context is three sentences you already know.
• Long one liners where the flags are the easy part and the string argument is the slow part.
• Notes into a heredoc, a scratch file or a README while you still remember what you did.
• Anything you would otherwise put off writing because your hands are busy and the window is small.

How to do it

1. Put the cursor at the prompt.
2. Hold your shortcut. You choose which key in Settings during setup and can change it whenever you want.
3. Talk for as long as you need. There is no session limit.
4. Let go. The text arrives at the prompt in about a fifth of a second.
5. Read it, then press return yourself.

Nothing is executed for you. The words stop at the prompt, which is the correct behaviour in front of a shell and is not going to change.

your shortcut

Listening… 0:04

Hold it to talk, tap it to toggle, Escape cancels. Pick the key in Settings.

Making the names come out right

This is where dictating into a terminal succeeds or fails. A general speech model spells what it thinks it heard and what it thinks it heard is English, so tool names become ordinary words and your service names become something adjacent. Turn on the developer tools vocabulary pack when Zumbo asks what you dictate most and the everyday ones arrive correct: the runtimes, the build tools, the package managers, the common commands.

For the rest, and especially for anything internal to your company, use Teach. Select the word that came out wrong, type what you meant, press Teach, and it is heard your way from then on. The correction stays on your Mac and the list grows out of your work rather than being typed in up front.

Over ssh, in a container, on a server

None of that matters. Zumbo composes locally and types into whatever has focus, so the far end of the connection is irrelevant. It is also why nothing about your session is uploaded to be transcribed: the speech engine is inside the app, there is no account, and there is no network request in the path.

Related: dictation for developers, dictating into Claude Code, and dictating into Xcode.