Push-to-talk dictation for macOS

Dictation that types for you.

Hold a key and talk like a person — false starts, ums and all.
Vocito types the sentence you meant. Punctuated, polished, yours.

fnRequest an invite How it stays private ↓

How it feels

Hold. Speak. It’s written.

No wake word, no floating orb, no editing pass. One held key, one breath, one clean sentence where your cursor was.

fn

Hold.

Like a walkie-talkie. Vocito hears nothing until the key is down, and stops the instant you let go.

Speak.

At the speed of thought. Trip over yourself, change your mind mid-sentence, say “um” — it’s all signal.

On my way.

It’s written.

Release the key. The polished sentence is already sitting in the field you were focused on.

key-up → polished text 0.00 s
median ~1 s across the beta fleetillustrative replay

Context-aware polish

Same breath. Right register.

Vocito sees where the words are landing — a mail draft, a Slack thread, a code file — and writes for the room.

“okay so um tell dana i’ll have the numbers by uh friday morning latest”
Mail formal
To: Dana Okafor
Subject: Q3 numbers

Hi Dana,

I’ll have the numbers to you by Friday morning at the latest.

Best,
Alex

Slack casual
# q3-planning
AX
alex 2:14 PM

dana — numbers coming your way friday morning, latest

Editor comment
41def close_quarter(books):
42  # TODO(alex): get Dana the numbers by Friday a.m. — hard latest
43  reconcile(books)

Multilingual, done honestly

Your languages, kept honest.

Script-true output — no romanized mush, no surprise translations, punctuation native to each script.

EN

Keep the meeting on Thursday.

FR

On garde la réunion jeudi.

BN

মিটিং বৃহস্পতিবারেই থাকছে।

DE

Das Meeting bleibt am Donnerstag.

English Français বাংলা everything else — off

Detection is restricted to the languages you enable — usually two or three. A borrowed word won’t derail the sentence into the wrong language.

Privacy as architecture

Your voice never leaves our metal.

Not a policy promise — a network diagram. No third-party AI ever touches your voice.

  • Self-operated inference

    Our models on our GPUs in our racks. No OpenAI, no Google — no third-party AI API anywhere in the path.

  • Audio lives in memory

    Decoded on arrival, gone after transcription. Never written to disk — ours or anyone else’s.

  • Keys that die on command

    Encrypted in transit, end to end. Every Mac holds its own revocable credential — hashed at rest server-side, killable from your dashboard in one click.

Full security design doc available to beta members on request.

For people who live in text fields

Built for keyboards-first people.

Vocito behaves like part of macOS, because it types like you do — straight into the focused field.

Types where your cursor is

Native insertion in every app — no plugins, no paste dance. If you can type there, Vocito can.

MailSlackXcodeTerminalNotionanywhere

A dictionary that’s yours

Teach it your names, jargon and acronyms once. Spelled right forever, on every device.

“anais”⇢Anaïs
“cube control”⇢kubectl
“q three okrs”⇢Q3 OKRs

Your key, your rules

Pick the push-to-talk key. Hold to talk, release to stop — the mic is physically idle any other time.

fnright ⌘right ⌥⌥ space

Linking without typed secrets

A new Mac shows a short code; you approve it in your browser. Single-use, expires on its own, revocable any time. Your dictionary follows — no password crosses your keyboard.

T2P·MK4single use

Private beta

Invites go out in small batches.

We’d rather be excellent for a few than adequate for everyone. Leave an email — we onboard weekly.

You’re in line. Watch your inbox.

macOS 14+ · Apple silicon & Intel · no card, no spam