Voice input for every field in Cursor, not just chat.
Talking to the agent is the part Cursor already solved. The commit message, the comment, the rename dialog and the terminal pane are the parts you still type.
The gap
Cursor has voice input for its agent. It is genuinely useful and this page is not an argument against it. The limit is scope: it is built for one panel. The moment you leave the chat box you are back on the keyboard, and the keyboard is where the rest of the day is.
Count what you actually type in an hour of work in Cursor. A prompt or two. A commit message. A comment explaining why the obvious approach did not work. A pull request description. A branch name. A message to a colleague about what you just found. One of those is covered.
What system wide dictation covers instead
Zumbo does not integrate with Cursor. It types where the cursor is, which means it works in every one of these without knowing they exist:
Hold it to talk, tap it to toggle, Escape cancels. Pick the key in Settings.
- The editor itself, for comments and documentation strings.
- The commit message box in the source control panel.
- The integrated terminal, including a coding agent running in it.
- Search and replace, rename dialogs and the command palette.
- The agent chat box as well, if you prefer one shortcut for everything.
How to use it
- Put the cursor where the words should go.
- Hold your shortcut. You pick which key in Settings during setup and can change it any time.
- Talk at your normal speed, for as long as you need. There is no session limit.
- Let go. The text lands in about a fifth of a second, punctuated.
Making it spell your stack
A dictated commit message with three misspelled package names is worse than a short one. Turn on the developer tools vocabulary pack and the everyday names arrive correct: runtimes, build tools, package managers, common libraries. Internal service names are not in anybody's model, so 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 it accumulates into a vocabulary that is specific to your codebase.
The commit message test
Here is the honest way to decide whether this is worth fifteen dollars. Look at your last twenty commit messages. If most of them are under eight words, you are not writing short commits because you like them, you are writing them because typing the real explanation is slower than the explanation is worth. Say the real one out loud and time it. It is about eleven seconds.
Related: dictation for developers, dictating into Claude Code, and voice typing in VS Code.
Does Cursor already have voice input?
Cursor has voice input for talking to the agent. That covers the chat box. It does not put your voice into a file, a rename dialog, a commit message field or a terminal pane, which is where most of the typing in a day actually happens.
So what does this add?
Dictation everywhere in the window rather than in one panel. The same shortcut works in the editor, the search field, the commit box, the integrated terminal and every dialog, because it types where the cursor is instead of integrating with one feature.
Will it spell my dependencies correctly?
Turn on the developer tools vocabulary pack and the common names come out right the first time. For your own internal names, correct a word once with Teach and it is heard your way from then on.
Can I dictate code itself?
You can, but it is rarely the right tool. Dictation earns its place on the prose around the code: prompts, commit messages, comments, pull request descriptions, and the notes you would otherwise not write.
Does anything leave my machine?
Not from Zumbo. The speech engine is inside the app and composes on your Mac. What Cursor sends is Cursor's business and unchanged by this.
Free for 3 days, everything unlocked. Then $15 once. macOS 14 or later, Apple silicon.