The fastest way to work with Claude Code isn't to type — it's to talk. Set Clavio's wake word to “Claude” and just say “Claude, add a loading state and handle the error case.” The words drop into the prompt and send themselves — nothing to press, hold, or click. Turn cleanup off and the model gets exactly what you said, the instant you finish. The same voice writes your commit, your pull request, your Slack reply — the whole coding loop becomes something you say, not type.
Talk to it by name: make the wake word “Claude”
Clavio lets you choose a wake word for each app, so for Claude Code you set it to the model's own name — “Claude.” Now you're not firing up a dictation tool; you're speaking to the model. You say “Claude, split this file into smaller components,” it answers to its name, types your instruction into the prompt, and — with auto-send on — sends it the moment you stop talking. No key to hold, no button, no reaching for the trackpad — just your voice.
Fast mode: your exact words, the instant you finish
A prompt isn't an email — it shouldn't be smoothed over or reworded. So for Claude Code, keep Clavio on Fast: it types your words exactly as you said them, with no rewriting and no processing pause, the second you stop. That's what you want for coding — the model gets your literal instruction at the speed of speech. Add your stack to Clavio's dictionary — framework names, CLI tools, your repo and service names — and “useEffect,” “kubectl,” and “clavio-backend” come through clean every time.
Watching your tokens? Let Clavio tighten the prompt
Long agent sessions add up, and if you'd rather send less, Clavio can do the opposite of Fast. Give Claude Code a polish that doesn't dress your words up but trims them down: you say the idea loosely, and a short, clean instruction is what reaches the model. Fast for raw speed, a tightening polish to spend fewer tokens — both are just a per-app setting, and you pick whichever fits how you work.
One voice for the whole loop — commits, PRs, Slack
The prompt is only the start. When Claude Code finishes a change, dictate the commit message straight into your git app. Speak the pull request — what changed, why, how to test — and here you may want it polished, so Clavio switches to a natural tone for that window while Claude Code stays on Fast. A reviewer pings you on Slack; you switch over and reply in a casual voice. Clavio remembers the tone for each app, so one wake word carries you through the entire loop — and since it only answers to your voice, a colleague talking nearby never sets it off. That's what hands-free coding actually feels like: you talk, and everything gets written.
Recommended Clavio settings for Claude Code
| Setting | Recommended | Why |
|---|---|---|
| Wake word (per app) | Make it “Claude” | Speak to the model by name — “Claude, refactor this” — instead of a generic trigger, with no key to hold. |
| Polish level | Fast (off) | Types your words instantly, exactly as said. No rewriting, no processing pause — the model gets your literal instruction. |
| Auto-send | On | Fully hands-free: the moment you stop talking, the prompt sends itself. You just speak — no key at all. |
| Personal dictionary | Add your stack | Keeps framework, CLI, and repo names — useEffect, kubectl, clavio-backend — coming through clean. |
| Token budget (optional) | A tightening polish | Instead of Fast, a polish that trims your words down, so fewer tokens reach the model. |
Code hands-free with Clavio
Clavio is an AI dictation app for Mac. It sits in your menu bar and types finished text wherever your cursor is — the Claude Code prompt, your git app, GitHub, Slack, the browser. You set a wake word per app (say “Claude” to talk to Claude Code), and choose the polish per app too: Fast with no cleanup for instant, exact prompts, or a tightening polish to save tokens. A personal dictionary keeps your stack's names intact, and an on-device voiceprint means only your voice sets it off. Free to try: 3,000 words a month, no card, macOS 14+ on Apple Silicon. Pro is £12/month for unlimited.
Keep reading
Common questions
Does Claude Code already have voice input?
It has a built-in voice option you start from the terminal and hold a key to use. Clavio is different in two ways that matter day to day: you give it a wake word (say “Claude” — nothing to hold), and it works everywhere your work lives, not just the terminal — your commit message, your PR, your Slack reply. That reach is why most people reach for it.
Can I dictate to Claude Code fully hands-free?
Yes. Give Claude Code the wake word “Claude,” turn auto-send on, and just say “Claude, do this” — the prompt is typed and sent the moment you stop talking. No key, no button — just your voice.
Will it reword my prompts or mangle technical terms?
Not on Fast — it types your words exactly, no rewriting. To keep framework, CLI, and repo names precise, add your stack to the dictionary. Want it shorter instead? Switch on a tightening polish — but by default Fast hands the model your exact wording.
Can dictation help me save tokens?
Yes, with the opposite setting. Give Claude Code a polish that compresses your speech into one short, clear instruction: you talk loosely, a compact prompt reaches the model. Fast is fastest and most exact; a tightening polish spends one round-trip to cut the length. You pick per what matters — speed or tokens.
Can I use the same voice for commits, PRs, and Slack?
Yes — that's the main reason to use a system-wide app instead of the terminal's own voice. Clavio types into your git app, GitHub, and Slack the same way it types into the Claude Code prompt, and remembers the tone per app: exact for the prompt, tidy for a PR description, casual for a chat reply. One wake word covers the whole loop.