unmute does two things: it types what you say, and it does what you say.

A Mac dictation app that also, casually, runs your computer — and then keeps track of the work while it runs. Hold a key. Talk. Typing was a mistake; we're walking it back.

// we know our landing page is simple, but hey — this is the best way to convey the message without the noise. so read it, and don't complain.

// the first key — dictation, except fast

You've dictated before. Not like this. Under a second to text, always — 10 seconds of audio or 10 minutes, same sub-1-second latency, every single time. We're not going to explain dictation to you; you know what it is. We're going to out-run everyone who's ever sold it to you. The speed is the whole pitch.

And your words do not die. Wi-Fi quits mid-sentence — keeps going. Café internet from a dial-up fever dream — keeps going. The instant the network wobbles, dictation falls back on-device, automatically — no toggle, no spinner, no "something went wrong." Wispr Flow couldn't keep a sentence alive through a hiccup. So we had to.

What you copy while you're talking doesn't vanish either. Grab an error line, a URL, a screenshot, mid-sentence — the scratchpad holds it and composes it into what gets pasted, instead of letting the clipboard eat it. It only arms when you say so, so nothing gets swept up behind your back.

// raw by default — what dictation is actually for

Be honest: most of what you dictate goes straight into ChatGPT, Claude, Cursor. Prompts. And prompts don't need commas. So we paste raw — no auto-punctuation, no autocorrect editing you into a stranger.

Need it shaped? Select the text, hold Caps Lock, and say the format out loud — "three bullets," "less corporate," "make it sound human." You name it; we don't pick for you. Two seconds, done.

// the second key — remote · a voice remote for your coding agent

Dictation types it. Remote does it.

Hold right ⌥, say the thing, it happens — on your actual Mac.

The twist: unmute is dumb. Proudly, aggressively dumb. We didn't train a model — the smartest one already exists, you're already paying for it, and ours would only get in the way. We gave it ears. That's the innovation.

Which is the entire trust story. Remote is Claude Code — your own terminal session, just voice-driven. The best agent anyone has shipped, minus the part where you type at it. Nothing new to trust. Friendly neighborhood Claude Code. (Codex too, if that's your executor. We sit above it, not inside it.)

We got this wrong once and it's worth telling you how. unmute used to rewrite the session it started for you — a wall of our scaffolding pasted in as though you'd typed it, our model pinned over your choice, a hook blocking the end of every turn. Two people told us their terminal was full of our furniture and that Claude Code worked worse under us. Both were right. All of it is deleted. Remote now starts an ordinary session and watches from the outside — what's in your terminal is yours.

Here's the thing: you probably use Claude Code hard — for code, for a couple of workflows you'd defend with your life. And that's still maybe 20% of what it can actually do. Press a key, say it, watch the other 80% show up:

None of these are features we built. It's just your agent, finally within reach. The bottleneck was never the model — it was the keyboard.

The ones that touch apps and browsers go through our own hands — unmute drives them in the background, so nothing snatches your screen while you're mid-thought.

// the third key — the unmute agent

Hold right ⌘ and you're not talking to a session. You're talking to unmute.

It deliberately has no card of its own. It's the thing you talk to about your work, not another piece of work to manage.

// when it's more than one thing at once

Once you can start work by talking, you start more of it than you can hold in your head. So the question stops being "what's running" and becomes whose move is it.

Every session is, at any moment, either our move — it's working, leave it alone — or your move: blocked, failed, or just finished and waiting on you. That's the only thing unmute sorts by. Not newest. Not loudest. Things move down as they stop needing you, they don't fall off a cliff.

All of it renders in the notch — the menubar row and the display cutout, which turns out to be the right amount of screen for this. A pill while you're talking. A wall of what's waiting. The task itself. A pocket of what you've actually been working in.

Attention gets pulled, never grabbed. Nothing modal, nothing steals focus. When something becomes yours, you get one calm spoken line — and then it stands down, instead of parking a sentence over your screen for the rest of the afternoon.

// meetings — new

Double-tap left ⌃. It taps system audio and your mic, and hands you notes at the end.

The notes are written by your own connected CLI — never a cloud model. We had one in the fallback path and pulled it out before shipping: a third vendor quietly receiving the full text of your meetings isn't a trade we were willing to make on your behalf. The notes are the product; the transcript stays out of your way.

// this one is genuinely new, and we'd rather you heard that from us than found out.

// the flywheel (this is the actual pitch)

Agents get better the more you feed them — memory, project files, notes-to-self, the shape of how you work. Give one structure and it serves you better, permanently. The power users know this; they're forever curating context.

You're probably not. Not because you're lazy — okay, a little — but because doing all that by hand, in a terminal, is a chore, so the flywheel never starts spinning.

unmute spins it. "Remember I always want X." "What did I say about that last week?" "Start a task for this." Said out loud, in passing, zero friction. You end up feeding your agent without trying — and the more you feed it, the more it's worth to you. The value compounds somewhere you've genuinely never taken it before.

Honest notes: it needs a Claude Code Pro or Max plan, because it is one and it uses yours. It's as fast as Claude Code (not instant) and as reliable as Claude Code (the most reliable agent there is). And every time Claude Code levels up, so does unmute — free. how remote works

// price

Yearly is cheaper, obviously. Arithmetic still works in 2026. see pricing

Get unmute  // we allowed ourselves exactly one button. Restraint.

// macOS says "damaged"? It's signed & notarized — that's just a browser quarantine flag. Drag unmute into Applications, then run xattr -dr com.apple.quarantine /Applications/unmute.app once. more

// talk to a human — 15 minutes

If you live in Claude Code, Codex or Cursor, I'd like 15 minutes of your time.

Not a demo, not a pitch. There's nothing I could sell you on a call that this page hasn't already tried. I want to hear how you actually work — what you run all day, where it goes slow, and which part of it you'd hand to your voice if you trusted it enough.

Because the next piece is the hard one: you say a thing, an agent task starts. One gets stuck, you unblock it by talking. No terminal. I'd rather build that out of ten real conversations than a hundred confident guesses.

book 15 minutes  // pick a slot — that's the entire form. no phone number, no "when works for you?", no follow-up sequence.

// and yes, still exactly one button on this site. that was a link. if no slot works, dm me instead.