Voice coding: talk to your coding agent

Say what you want to Cursor, Claude Code or Codex instead of typing it. Hold Fn, explain the change the way you would to a colleague, let go and paste. The agent writes the code.

$2.99 once. Speech stays on your Mac. Apple Silicon, macOS 26 Tahoe.

Terminal
✻ Welcome back. /help for commands
cwd: ~/northwind-app · main
> Add retry with exponential backoff to the upload handler, keep the progress bar from jumping backwards, then run the tests.
Readsrc/upload/handler.ts
Edithandler.ts +18 −4
EditProgressBar.tsx +3 −1
Bashnpm test 42 passed
Done. Retries back off 0.5s → 8s and progress only moves forward.
>
fnHold to talk

Works in every agent you already use

Whisperer writes your words to the clipboard and you paste them with ⌘V, so it works in any chat box, terminal or editor, with nothing to install in them.

  • Cursor Cursor
  • Claude Code Claude Code
  • Codex Codex
  • Windsurf Windsurf
  • VS Code VS Code
  • Claude Claude
  • ChatGPT ChatGPT
  • Terminal, iTerm2, Warp

Why voice suits AI coding

Dictating syntax word by word was always slow. Describing intent to an agent is what speech does best.

Prompts are prose
An agent wants a clear paragraph about what to change and why. That is exactly what speech is good at.
More context, less effort
Saying the edge cases out loud is cheaper than typing them, so the agent gets a better brief and asks fewer follow-ups.
Your hands get a break
Prompts, PR descriptions, review comments and Slack replies are most of a day's typing. They can all be spoken.
Your code stays private
Speech becomes text on your Mac. No speech server hears what you say about an unreleased product.

Set up in two minutes

No plugin for your editor, no account, no API key.

  1. 1
    Get Whisperer

    Download it from the Mac App Store for $2.99 and allow the microphone.

  2. 2
    Let the model download

    The default speech model is about 547 MB and arrives the first time you dictate.

  3. 3
    Hold Fn and talk

    Describe the change. Words appear in the preview as you speak.

  4. 4
    Paste with ⌘V

    Let go of Fn, click into the agent's prompt and paste. Edit, then send.

Spells your stack the way you do

Speech models guess at names they have not heard: Kubernetes becomes “cooper netties”. Fix it once and Whisperer learns it, so the next prompt is right.

Whisperer
Dictionary
Learn from my corrections
Next dictation

Scale the Kubernetes nodes before the launch review.

Scale the Kubernetes nodes before the launch review.
YOUR WORDS
Kubernetesheard as “cooper netties”Learned
Gleanheard as “glean, clean”
Priya Shahheard as “Pria Shaw”
Northwindheard as “north wind”

Let the agent read your meetings too

With Pro, Whisperer runs a small MCP server on your Mac. Claude Code, Cursor and Codex can search your meetings and dictations, so “build what we agreed in Tuesday's review” has something to work from.

Claude
What did we decide in Tuesday's launch review? Make me a checklist.
whisperersearch_meetings“launch review”
whispererget_meeting“Mobile launch review”
Claude

From your Whisperer notes for Mobile launch review (Tue 11 Nov):

  • Launch moved to 24 November
  • Priya rewrites the first three onboarding screens by Thursday
  • Dan cuts the account step from onboarding before the 24th
  • The press embargo stays where it is
Reply to Claude…

What Whisperer does not do

Whisperer writes sentences, not syntax. There are no spoken-symbol commands like “open paren” and no casing commands like “camel case”. If you want to dictate raw code character by character, a dedicated voice-control tool fits better. If you want to tell an agent what to build, this is the faster path.

It also never types into apps for you. When you let go of Fn the text is copied and a Copied ⌘V toast appears, so you choose where it goes. Read more in how to dictate code on a Mac.

Voice coding questions

Say it. Paste it. Let the agent build it.

Whisperer is $2.99 once. Pro adds meeting notes and the MCP server for $14.99 once. No subscription.