You're on.
Just talk.

Ramble. Backtrack. Think out loud. SayYo turns it into clean text that's ready to send — and the more you use it, the more it sounds like you instead of like AI.

Free to start · Works in your browser · Nothing it learns leaves your device

What actually happens

Dictation gives you a transcript.
You wanted the message.

Every voice keyboard can write down what you said — every "um", every false start, every correction. Then you fix it yourself. SayYo skips that step: it reads what you meant and writes the version you'd have typed if you had the time.

HEARD “um so hey it's — it's me, can you can you tell mike that the the meeting it moved to tuesday, no wait, thursday, thursday at 2, and uh yeah tell him thanks”

Same twelve seconds of talking. Pick the shape it should land in.

The part that compounds

Every AI writes the same way.
You don't.

You can spot it now, can't you — "I hope this finds you well", the stacked em-dashes, the three-item list where you gave two. Every dictation app polishes your voice into that. SayYo does the opposite, and it learns the difference from the one signal nobody else keeps: the edits you already make before you hit send.

SayYo wrote Hi Mike, I hope this finds you well. Quick heads-up — the meeting has moved to Thursday at 2:00 PM. Thanks, Alex
You send Hey Mike — meeting moved to Thursday at 2. Thanks!

↑ Rewrite that line however you'd really send it

That bottom panel isn't a mock-up — it's SayYo's actual module reading your edit, running in your browser, sent nowhere. It assumes two earlier edits alongside yours, because one edit is a mood and three is a habit; SayYo waits before it names a pattern.

In a hurry? Paste a few things you've already written — or screenshots of a chat thread — and it picks up your voice on day one, before you've corrected anything.

Built around the awkward parts

Speaking is fast. Sounding good is the slow bit.

It gets the names right

Dictation mangles proper nouns — every time, forever. Fix a colleague's name once and SayYo remembers the spelling, then feeds it back so it stops mishearing them.

It answers the actual thread

Paste what you're replying to, then talk the way you'd really talk — "yeah, Tuesday works, but not before ten". It fills in what you left unsaid, and never answers what you didn't.

Shortcuts you say

Say "my calendar link" and get the link — exactly. For the addresses, booking links and account numbers dictation always mangles and you always fix.

It knows the channel

An email gets a greeting and a sign-off. A text gets one line. A task gets a checkbox. Same words in, right shape out.

Fix it by talking

Not quite right? Say "make it shorter" or "warmer, and mention Friday". No cursor, no retyping.

Speak one language, send another

Talk in the language that comes naturally and send polished text in any of 13 others.

Where it all lives

It learns you. The learning never leaves your device.

An app that studies how you write is only worth having if you can see what it thinks and delete it. Everything SayYo learns about you is stored on your own device — not in an account, not on our servers, not in anyone's training set.

  • Your style profile, your learned words, your history — all local. Nothing syncs.
  • Readable and editable. Open Settings and you can see the words it has picked up and the description it has written of how you sound — and correct either.
  • One tap to forget all of it.
  • What does leave: your recording and the words in it, sent to OpenAI through our proxy to be transcribed and written up. Nothing else, and nothing is kept there to train on.
Coming to iPhone

A keyboard, so it works where you already type.

Messages, Mail, Slack, wherever — tap the key, say your piece, drop the clean version straight into the field. In testing now; the web app works today on any phone.

And it won't ask for "Full Access" — the permission that lets a keyboard read everything you type, including your passwords. Other voice keyboards require it. Ours is built so it doesn't need it.

Stop editing your own voice notes.

Try it once, on something you actually need to send. It starts sounding like you from the first thing you change.

No card. No install. Takes about twelve seconds.