Speaky Box for
Claude
ChatGPT
Cursor
Slack
Mail
Google Docs
Microsoft Word
Apple NotesCompared with
Read
Hold a key, explain what you want in plain English, and Speaky Box types a task Cursor's agent can act on: what to build, where, the constraints, and what done looks like.
Cursor does its best work with a clear task, the constraints, and a definition of done. Speaky Box writes that brief from what you said, so you spend the typing time reviewing the diff instead.
Add your file names, libraries and internal names to Speaky Box once and it spells them correctly every time, in every app.
Claude Code, Codex, Warp, Ghostty. If you can type in it, you can talk in it.
Yes. Cursor has built-in voice input for its agent. Cursor 2.0, released on 29 October 2025, added Voice Mode, which Cursor describes as controlling the agent with your voice using built-in speech-to-text. In the Agents Window, Cursor 3.1 (13 April 2026) lets you press and hold Ctrl+M to speak.
Cursor's 3.1 changelog says voice input now records your full voice clip and transcribes it in one batch for better accuracy. While you record, Cursor shows a waveform, a timer, and buttons to cancel or confirm. Cursor's Design Mode in the Cursor browser also takes voice: you can describe a UI change out loud, and the mic stays available while an agent is mid-run so you can queue the next change.
Sources: Cursor changelog 2.0, 3.1 and Design Mode Improvements (5 June 2026), read 17 October 2026.
To use voice input in Cursor's Agents Window, update Cursor to 3.1 or later, click into the agent input, then press and hold Ctrl+M while you speak. Cursor records the whole clip, shows a waveform and timer, and turns it into text when you confirm. Read the text, then send it.
Ctrl+M shortcut and the recording controls arrived in 3.1.Ctrl+M and say your request.Cursor Voice Mode supports submit keywords. Cursor's 2.0 changelog says you can define custom submit keywords in settings, so saying the chosen word sends your spoken request to the agent without a click.
Sources: Cursor changelog 2.0 and 3.1, read 17 October 2026.
When Cursor voice input does nothing on a Mac, check three things first: that Cursor has microphone access in System Settings > Privacy & Security > Microphone, that the right microphone is picked in System Settings > Sound > Input, and that Cursor is up to date.
Sources: Apple's Control access to the microphone on Mac and Change your sound input settings, and Cursor's 3.1 changelog, read 17 October 2026.
Cursor does not publish a list of languages for its voice input, as far as we could find in its docs and changelog. What Cursor's release notes do say: from 24 September 2026 (version 3.22), voice submit keywords work in any language, and in the Agents Window, submit keywords in non-Latin scripts such as Cyrillic or Chinese trigger auto-submit while you dictate.
If you use macOS Dictation instead, set the language in System Settings > Keyboard > Dictation: click Edit next to Languages and pick a language and region.
Sources: Cursor release notes and Apple's Use Dictation guide, read 17 October 2026.
macOS Dictation is free and works in any Cursor text field, including the agent input, the editor and the terminal. Turn it on in System Settings > Keyboard, under Dictation. Click where you want the text, then press the Microphone key, your Dictation shortcut, or choose Edit > Start Dictation.
In supported languages, macOS Dictation adds commas, full stops and question marks for you. Apple says you can dictate text of any length, and Dictation stops on its own after 30 seconds with no speech. You can pick the microphone under Microphone source in the same Dictation settings.
What macOS Dictation gives Cursor is a transcript. Every "um, no, the other file" goes into the prompt as you said it. Source: Apple's Use Dictation guide, read 17 October 2026. More on Speaky Box vs Mac dictation.
We make Speaky Box, so weigh this table with that in mind. It covers writing prompts and code changes in Cursor on a Mac.
| Cursor voice input | macOS Dictation | Speaky Box | |
|---|---|---|---|
| Where it works | Cursor's agent input and Design Mode | Any text field in Cursor and other Mac apps | Wherever your cursor is: Cursor's agent box, the editor, the terminal, and every other Mac app |
| What lands in Cursor | A transcript of the clip, after you confirm | A transcript of what you said | Finished text, or with the Smart key, a structured prompt suited to the app in front |
| Ums and false starts | Not stated by Cursor | Kept | Removed |
| Custom terms | Not stated by Cursor | Not covered in Apple's Dictation guide | A Words list for function names, library names and file names |
| Cost | Part of Cursor | Free, built into macOS | Free for 2,000 words a week, Pro £8.99 a month |
If your requests to Cursor are a sentence or two, Cursor's own voice input or macOS Dictation will do, and neither costs extra. Speaky Box is worth it when one spoken request to Cursor runs long: it strips the ums and lays the prompt out as task, files, constraints and what done looks like. For a full routine, see our guide to voice coding in Cursor.
Sources: Cursor changelog 3.1, Apple's Use Dictation guide and our pricing page, read 17 October 2026.
src/auth/session.ts, the refreshToken function", not "that auth bit". Add names you use often to the Speaky Box Words list.session.test.ts pass." Our four-part prompt structure covers this.Dictating to other AI tools works the same way. See how to prompt AI by voice, Claude Code voice dictation, and Speaky Box for Claude and ChatGPT.