Dictate Into Cursor on Mac: Voice Prompts for the Agent

Cursor already has voice input in its agent chat: click the microphone, or hold Ctrl+M in the Agents Window (Cursor docs, checked October 21, 2026). Sorimac is for everything else on the screen: the code editor, the integrated terminal, commit messages and other apps, transcribed on the Mac. It costs $9.99 once.

How do you set it up?

Install and allow access. Download Sorimac, then allow Microphone and Accessibility when macOS asks. Accessibility is what lets it paste into Cursor.
Pick when it listens. In Sorimac settings choose Push-to-Talk if you talk to colleagues while coding, or Always On if you work alone. Leave Auto Return on for the agent chat, or turn it off if you want to read each prompt first.
Click into Cursor and speak. Put the cursor in the agent chat, the editor or the terminal panel and say what you want. The text lands where the cursor is.

What do people say into Cursor?

“Refactor the fetch user function to use async await and keep the error messages the same”
“Write a test for the date parser that covers leap years and an empty string”
“Explain why this component re-renders every time the parent updates”

What to watch for in Cursor

Auto Return sends the prompt

Auto Return is on by default with a one second delay: after the text lands, Sorimac presses Return once things go quiet. In the agent chat that submits the prompt. In the code editor it inserts a new line, which you may not want. Set it to Off in settings if you mostly dictate into the editor.

Prompts, not code

Sorimac types the words you say. Speaking works best for the instructions you give the agent; for exact syntax, let the agent write the code.

Voice commands start with the wake word

Plain speech is typed. Only sentences that start with the wake word, such as "Computer, switch to Safari", are treated as commands, so describing a bug will not trigger anything.

Your clipboard

Sorimac pastes through the clipboard: it saves what was there, pastes with Cmd+V, and puts the old text back about a second later. A clipboard manager will usually record each dictation as a new entry. Only text is restored, so a copied image is replaced.

Cursor's own voice input vs Sorimac

Cursor's docs say to click the microphone in the chat input and review the transcription before sending. Cursor 3.1 (April 13, 2026) moved the Agents Window to batch speech-to-text and added press-and-hold Ctrl+M with a waveform, timer and cancel or confirm buttons. If you only ever talk to the agent, that may be all you need.

Cursor's pages we read do not say whether its transcription runs on your Mac or in the cloud, so we do not claim either. Sorimac transcribes on the Mac and works in any text field, in Cursor and outside it.

Sources, checked October 21, 2026: Cursor docs: Prompting agents, Cursor 3.1 changelog (April 13, 2026).

FAQ

Does Cursor have voice input?

Yes. The agent chat has a microphone button, and the Agents Window supports holding Ctrl+M to speak (Cursor docs and changelog, checked October 21, 2026).

Can Sorimac type into the Cursor code editor?

Yes. It pastes at the cursor in any text field, including the editor and the integrated terminal. Turn Auto Return off if you do not want a new line after each sentence.

Does my voice leave the Mac with Sorimac?

No. Speech recognition runs on the Mac. Cloud transcription is only used if you add your own API key and turn it on.

Will a sentence about code trigger a voice command?

No. Commands need the wake word at the start, such as "Computer". Everything else is typed.

Dictate into Cursor on Mac