How to Dictate Into Claude Code (and Cursor) by Voice on Windows

To dictate into Claude Code on Windows, install PipeVoice, a free push-to-talk voice typing tool, then hold Ctrl+\, speak your prompt, and release. PipeVoice types real keystrokes straight into the Claude Code CLI, your terminal, Cursor, or any focused app, so you are not limited to a single editor's built-in dictation.

Why dictating prompts beats typing them into AI coding tools

Prompts for AI coding tools are long. You are not typing git status, you are describing intent: "refactor this handler to stream the response, keep the existing error shape, and add a test for the empty case." That is a sentence, and sentences are faster to say than to type.

Dictation also keeps you in flow. You can keep your eyes on the diff while you talk through the next instruction, instead of breaking concentration to hunt for keys. For anyone with RSI, or anyone who simply thinks out loud, speaking the prompt is less friction. And because spoken prompts tend to be more complete (you naturally add the constraints you would skip while typing), the model often needs fewer follow-up corrections.

Claude Code's built-in /voice vs a system-wide dictation tool

Claude Code has a /voice command, but it only types inside the Claude Code CLI. The moment you want to dictate into Cursor, a VS Code chat box, a browser, or a plain terminal, it cannot help you. A system-wide tool solves the whole desk, not one window.

PipeVoice sits at the operating system level and types into whatever app is focused. The same hotkey works in the Claude Code terminal, in Cursor's composer, in a commit message, in a GitHub issue, and in Slack. You learn one workflow and use it everywhere.

Setting up PipeVoice to type into the Claude Code CLI and terminal

Getting running takes a couple of minutes.

1. Download the installer: Pipevoice-Setup.exe.
2. Run it. PipeVoice is currently unsigned, so Windows SmartScreen may show an "unrecognised app" warning. Click More info, then Run anyway. (Code signing is in progress.)
3. Pick a transcription engine. For fast streaming where words appear live as you speak, choose Deepgram (bring your own free API key, roughly pennies a day). For maximum accuracy, choose OpenAI Whisper with your OpenAI key. For zero cost and full offline use, choose local Whisper / faster-whisper (first run downloads a roughly 150MB model).
4. Open your terminal with Claude Code running, hold Ctrl+\ (or Right Ctrl), speak your prompt, and release. The text types into the CLI as if you had typed it.

Dictating into Cursor, VS Code, and any chat box with one hotkey

Because PipeVoice emits real keystrokes into the focused app, the same hotkey works everywhere. Click into Cursor's composer and dictate a multi-line instruction. Click into the VS Code chat panel and describe the change. Click into a browser chat box and ask a question. There is nothing app-specific to configure for the basics: focus the field, hold the key, talk.

This is the core advantage over editor-locked dictation. One muscle memory covers git commit messages, pull request descriptions, Jira tickets, and your AI prompts alike.

Using "send it" and auto-Enter to fire prompts hands-free

PipeVoice understands voice commands inside your speech: say "new line", "new paragraph", "tab key", or "scratch that" to fix a slip. To fire a prompt without touching the keyboard, end with "send it", which presses Enter for you.

If you want every dictation in a given app to submit automatically, turn on auto-Enter for that app in a per-app profile (covered next). That makes the Claude Code loop fully hands-free: hold the key, say the prompt, release, and it sends.

Per-app profiles: different engine and cleanup for your editor vs terminal

Your terminal and your editor want different behaviour, and PipeVoice lets you set that per app. A per-app profile can change the transcription engine, the AI cleanup, whether auto-Enter is on, and whether output is typed or pasted.

A practical setup looks like this:

• Claude Code terminal: auto-Enter on, so prompts submit the moment you finish.
• Cursor / VS Code composer: auto-Enter off, so you can review and edit a long instruction before sending.
• Code-heavy contexts: use vocabulary boosting so jargon and library names transcribe correctly.

For more on this, see per-app dictation profiles and dictating into the terminal on Windows.

Keeping prompts clean with Flow mode polish

Spoken prompts come with filler words and missing punctuation. PipeVoice's optional Flow mode cleans filler, fixes punctuation, and corrects casing before the text lands. You choose the polish provider: OpenAI, Google Gemini (free tier), OpenRouter (free community models), or local Ollama for offline cleanup. Importantly, polish sends text only, never audio.

For non-native speakers, there is an accent and language picker (British, US, Australian, Indian, and New Zealand English, and more) plus a free-text "speech notes" field where you can describe your accent, a stutter, or heavy fillers so the cleanup handles them better.

Offline option: dictate prompts with nothing leaving your machine

If your code is sensitive, you can run the whole pipeline locally. Pair local Whisper for transcription with Ollama for Flow mode polish, and nothing leaves your PC: no account, no telemetry, no servers of ours. The trade-off is honest: local Whisper is slower than the cloud engines and wants a decent CPU for the larger, more accurate models.

If you do use a cloud engine instead, audio goes only to the provider you chose, on your own key. Either way there is no PipeVoice account to create.

How PipeVoice compares to other voice tools

Most dictation tools fall into one of a few camps: paid cloud apps, expensive desktop suites, the built-in Windows option, or power-user scripting tools. PipeVoice stakes out a specific corner of that map.

The wedge is simple: PipeVoice is free, open source, Windows-native, has a true offline path, lets you bring your own engine and key, and types into any app including the terminal and Claude Code. See the full comparison with Wispr Flow or read about voice coding on Windows.