Voice dictation for Cursor
Cursor's chat box, Composer and editor are ordinary text fields, so FreeFlow dictates into all of them: hold ⌥, describe the change, release, and the prompt is at the cursor, transcribed on your Mac. Add your project's identifiers to the dictionary once and useEffect, kubectl and your repo's odd names come out spelled right.
Why dictate prompts at all
An agentic session is mostly prose: what to build, what went wrong, what to try next. Typing is the slow step. Speech runs about three and a half times faster than typing for most people, and a spoken prompt tends to carry the context you would have trimmed to save keystrokes — which is exactly what an AI editor needs more of. Cursor has its own voice input, but it only works inside Cursor; FreeFlow works in Cursor, the terminal beside it, the browser where you are reading docs, and Slack when you report back.
Setting it up for code
- Install FreeFlow and allow Microphone and AccessibilityDownload, drag to Applications, allow both prompts. The 14-day trial is the whole app.
- Make a “Code” modeModes → +. Parakeet V2 if you only ever speak English (6.1% word error rate, fastest), Parakeet V3 otherwise. Leave formatting off so an identifier is never “corrected” into a word; keep filler removal on.
- Teach the dictionary your identifiersThe Dictionary page in FreeFlow's window. Framework names, your services, your teammates.
useEffect,Postgres,kubectl, the product's code name. Add them as replacements (heard-as → spelling): those rewrite exactly on every model, including Parakeet, which does not take vocabulary hints yet. - Hold ⌥ in Cursor's chat boxSay the prompt, release. For long prompts use ⌥Space to toggle hands-free and press it again when you are done.
Where it works inside Cursor
- Chat and Composer — the main use. Dictate the whole prompt, then hit Enter yourself; FreeFlow never submits anything on your behalf.
- The editor — comments, docstrings, commit messages in the Source Control box. Dictating code itself is possible but rarely worth it; dictating what the code should do is.
- The integrated terminal — hold ⌥ and the command lands at the prompt. See the terminal page for the details.
- Any dialog with a text field — rename, search, the command palette.
What FreeFlow does not do here
It does not read your workspace. Some dictation tools match spoken file names against the open project; FreeFlow does not, and the dictionary is how you get the same spelling. It does not drive Cursor's agent — the voice cards that answer an agent's questions exist for Claude Code, through its hook system, and Cursor has no equivalent hook. And it does not submit: the prompt is pasted, you press Enter.
Questions people ask
Does it work with Cursor's agent mode?
Dictation into the agent's prompt box, yes. Answering the agent's questions by voice, no — that card system is specific to Claude Code's hooks.
Will it mangle camelCase and snake_case?
Out of the box it will often hear useEffect as “use effect”. One dictionary entry fixes each identifier permanently. A Code mode with formatting off keeps them untouched.
Can I dictate actual code?
You can, and a dictionary full of your identifiers helps. Most people find it faster to dictate the intent and let Cursor write the code.
Does Cursor have its own dictation?
Yes, inside Cursor only. FreeFlow's advantage is that the same key works in the terminal, the browser and Slack, with the same dictionary, and the audio never leaves the Mac.