Developers

Voice coding

Dictate your prompt and watch it land as text. Then send it to the AI agent that writes the code.

The idea

Talk to your IDE, let the AI build it

Voice coding means you describe what you want out loud and an AI agent turns it into code. You hold a shortcut, speak the prompt, and the text appears wherever your cursor is. The agent reads it and writes the implementation. You review, accept, move on.

Andrej Karpathy named this "vibe coding" in February 2025. He was talking to Cursor's Composer through Superwhisper, describing features in plain English, barely touching the keyboard. If you want the full story behind the term, the vibe coding page covers it.

Why it works

Speaking changes how you write prompts

When you type a prompt you tend to be terse. When you talk you give the agent more to work with: the edge cases, the why, the thing you tried that didn't work. Coding agents handle that verbose input fine, and the extra context often gets you a better result on the first pass.

There's a speed gap too. Most people talk around 3x faster than they type, which matters a lot when a good prompt runs a few paragraphs. You can measure your own gap with the words per minute test.

How it fits

Works in whatever editor you already use

Superwhisper runs at the operating system level. It listens for a global shortcut, records, transcribes, and drops the text at your cursor. Nothing to install inside your editor. If you can type in it, you can talk to it.

That covers Claude Code in your shell, Codex, OpenCode, plus Cursor Composer, Copilot Chat in VS Code, Windsurf, Xcode, and JetBrains. It also handles the writing around the code: commit messages, PR descriptions, the Slack note explaining a technical call.

Clean output

Super Mode keeps your prompts technical

Raw dictation is noisy. You say "um" and "wait, no" and the model has to guess where the sentence ends. Super Mode reads what's on your screen and formats around it. In a code context it keeps the text technical and tidy, so you get well-formed instructions instead of "hey comma can you please add a function that dot dot dot."

It runs fully on-device if you want it to. Audio never leaves your machine, which is what makes it usable on air-gapped setups and under SOC 2 Type II and HIPAA requirements. The on-device models also work on a plane with no internet.

Press your shortcut

A global hotkey starts recording in any app.

Describe the change

Talk through the feature, the bug, or the refactor.

Text lands at the cursor

Punctuated, formatted for the editor you're in.

The agent writes the code

Cursor, Claude Code, or Codex takes it from there.

Keep exploring

Support

Frequently asked questions

Start voice coding today

Free tier that doesn't expire. macOS, Windows, and iOS.

Download free