TL;DR: Voice typing for Claude Code means speaking your prompts instead of typing them, and it is the highest-leverage upgrade for agent-heavy work. Typing is the bottleneck: with Infina you can speak thousands of words of prompts a day, keep multiple Claude Code sessions busy, and ship faster. Infina runs the whole loop hands-free (dictate, send, switch apps) with on-device transcription, for $99 once with a 7-day refund. That one purchase pays for itself in reclaimed hours.

Infina is our product, so we are biased. The workflow below is still accurate, including the parts where a free tool is genuinely enough.

Why typing is the bottleneck in Claude Code

Claude Code turns coding into a conversation: you describe what you want, the agent does the work, you review and redirect. Your main physical activity is no longer writing code. It is writing prompts.

And good prompts are long. "Fix the bug" gets you a mediocre result; two or three sentences of goal, context, and constraints get you a good one. Then the agent replies, and you write another paragraph.

Most people speak roughly three times faster than they type. When your job for the day is producing a few thousand words of instructions, that gap is the whole ballgame:

  • Long prompts stop feeling expensive. You stop compressing your intent into terse commands that the agent then misreads.
  • Iteration gets cheap. "Actually, also handle the empty-state case and don't touch the config file" is a two-second utterance instead of a typing chore.
  • You stay in review mode. Your eyes stay on the agent's output while your voice queues the next instruction.

This is the productivity math in one line: speak more words per day, keep more agents busy, do more in less time. The same case, across Claude Code, Cursor, and Codex, is laid out in dictate prompts to AI.

Raw dictation is fine: Claude doesn't care about your commas

Here's the unlock that makes Claude Code dictation practical: LLMs understand unpolished speech. "uh so refactor the auth middleware to use the new session store and um make sure the tests still pass" produces the same result as a beautifully punctuated version.

This is the opposite of dictating an email. For human-facing text you need cleanup; for AI prompts you need speed and fidelity to your intent.

That's exactly why Infina's base product ships raw on-device dictation with fast rule-based cleanup instead of an LLM rewrite layer. For prompting, polish is latency you don't need. (If you also dictate emails and docs, that's what the optional cloud add-on is for; more in best dictation apps for Mac.)

Workflow 1: push-to-talk (the simple one)

This is where to start, and it is how many of our users dictate to Claude Code half the time:

  1. Click into your Claude Code terminal.
  2. Hold Option (⌥) and speak your prompt.
  3. Release Option: the text lands at the cursor.
  4. Press Enter to send.

That's it. It works in any terminal (Terminal.app, iTerm, Warp, the VS Code terminal) because Infina types into whatever app is focused, at the OS level.

Transcription runs on your Mac, on an on-device speech model on the Apple Neural Engine. It works offline and your audio never leaves your device.

Push-to-talk is the mature, reliable path. If all you want is to stop typing prompts, you can stop reading here.

Workflow 2: hands-free voice typing for Claude Code

This is the mode that changes how you work. Push-to-talk still tethers you to the keyboard: every prompt starts with your hand on Option and ends with your finger on Enter. Hands-free mode removes both.

  1. Double-tap Cmd (⌘) to turn hands-free mode on. Your Mac now listens; detection runs on-device, so nothing is recorded or sent anywhere while it waits.
  2. Speak a sentence that starts with "type", then your prompt. Infina types it into Claude Code.
  3. Say "send": Infina presses Enter for you.
  4. Need a different session? Say "open Terminal" (or any app by name) and keep going.

It works from two or three feet away: leaning back in your chair, standing at a whiteboard, pacing while you think. Double-tap Cmd again to turn it off.

One honest note: hands-free is our newest surface and still labeled experimental, so it ships off by default and a quiet-ish room helps. It is English-only in the base product, and push-to-talk is the fallback that always works. The full picture is in hands-free voice prompting.

Running multiple Claude Code agents by voice

The pattern that actually sells this to people who try it: parallel agents.

Say you have two Claude Code sessions going (one refactoring, one writing tests) plus Cursor open on a third task. Hands-free, the loop looks like:

  • Review agent 1's diff on screen. Say "type" plus the correction, then say "send".
  • Say "open Terminal" to land in agent 2's window. Dictate its next task, "send".
  • Glance at Cursor, queue an instruction there too: same loop, different app.

Your hands never leave your coffee. The bottleneck stops being your typing and becomes your judgment, which is where you want it.

This is how one person keeps several agents shipping at once. (The same loop works for Cursor; see voice typing for Cursor.)

How to speak a good prompt

Dictated prompts fail the same ways typed prompts fail: too vague, missing context. A structure that survives being spoken out loud:

  1. Goal first. "Add rate limiting to the public API endpoints."
  2. Context second. "We use Express, the middleware lives in src/middleware, and there's an existing Redis client."
  3. Constraints last. "Don't add new dependencies, and don't touch the auth middleware."

A few speech-specific tips:

  • Spell odd identifiers once. Say "the file is p-t-t handler dot t-s" the first time; after that, "the PTT handler" is unambiguous in context.
  • Don't restart the sentence when you stumble. Say "scratch that" and keep going. Claude parses the correction fine.
  • End with the ask. "Show me the plan before you change anything" as the last sentence gets respected more reliably than buried in the middle.

Our voice-command reference lives at infina.so/docs if you want the full list of what you can say.

Honest alternatives (including free ones)

Any dictation tool can do the typing half of this. Being straight about the options:

  • macOS built-in dictation: free, already on your Mac, and it types into a terminal just fine. Worth trying first. You'll still trigger it with a key and press Enter yourself, and accuracy on technical vocabulary is hit-and-miss.
  • Wispr Flow, Superwhisper, and similar: solid push-to-talk dictation apps; comparisons at Wispr Flow vs Infina and Superwhisper vs Infina. All of them stop at typing text: hotkey to trigger, and sending plus window-switching stays your job.
  • Infina: the only one on this list that runs the full loop. It triggers by voice, sends by voice, and switches apps by voice, with transcription on-device by default. It's $99 one-time (at the time of writing) with a 7-day no-questions refund; details on pricing.

If you dictate two prompts a day, use the free thing. If you prompt agents for hours, the loop is what you're buying, and $99 once pays for itself fast.

FAQ

Can I use free macOS dictation with Claude Code? Yes. It types into any terminal like any other text field, and it costs nothing. You'll trigger it from the keyboard, press Enter yourself, and live with weaker accuracy on technical terms, but it's the right first experiment.

Do I need to dictate punctuation for Claude Code prompts? No. Claude understands unpunctuated, conversational speech, filler words included. This is the big difference between dictating prompts and dictating prose for humans.

Can I send a prompt to Claude Code without touching the keyboard? With Infina's hands-free mode, yes: speak a sentence that starts with "type", then say "send". Infina types the text and presses Enter; mainstream dictation apps type the text but leave sending to you.

Does Infina's voice typing work offline? Yes. By default transcription runs entirely on your Mac (Apple Silicon required), so it works with no internet and your audio never leaves your device. See on-device dictation for Mac.

How accurate is dictation on technical terms? Infina's on-device model hits 95%+ accuracy for clear speech. Unusual names and jargon are the weak spot of any speech model; the optional $10/month cloud add-on uses larger cloud models that are sharper on names and jargon.

What does Infina cost? $99 one-time (price at the time of writing) with every 1.x update included and a 7-day money-back guarantee. No subscription and no free trial; the optional cloud add-on is $10/month.

The bottom line

Claude Code moved the work from writing code to writing instructions, and speaking instructions is simply faster than typing them, with zero penalty because the model doesn't need polished text.

Start free with macOS dictation to prove the habit. If you find yourself prompting agents all day, the hands-free loop (speak, send, switch, from across the room) is the version of this that compounds: more prompts per day, more agents in flight, more shipped.

That specific loop is what Infina was built for, at $99 once, risk-free for 7 days.