NewKoi for macOS

Speak naturally.
Send it as is.

Press a key, talk, and Koi types clean, ready-to-send text in any app on your Mac — fillers gone, changes of mind applied, formatted for wherever you're typing.

macOS 26 or later · Apple silicon · English, Chinese, Japanese and Spanish

Animated demo. In a chat, replying to “Got time for a call this week?” by saying “um, let's do tomorrow at three, no wait, four thirty” types “Let's do tomorrow at 4:30”. In an email, selecting a sentence and saying “make it more polite” rewrites only that sentence.

Filler dropped, correction applied, time written as digits. No period — it's a chat.

Two modes. One key.

Both start from the right Option key: tap or hold it to dictate, and add right Shift to edit.

Dictate

⌥right

Talk the way you talk. Koi writes what you meant to type.

TapHands-free — tap again to finish
HoldPush to talk — let go to finish
EscCancel the recording

Edit

⌥right+⇧right

Say what you want done to your text.

With text selected
Make it more formalFix the typosMake it shorter
With nothing selected
Write a quick thank-you to Tom for fixing the build

You can change both shortcuts in Settings and add an alternate for each.

Formatted for wherever you're typing.

Koi sees which app you're in and lays out the same words to fit: relaxed in chat, punctuated in email, a numbered list in a doc.

  • Terminal: always a single line — never a stray Return that runs a command.
  • Code editors: only the fillers come out. Identifiers, paths and commands stay exactly as said.
You said

um so three things before Friday first ship the beta second update the docs and third tell the team

Three things before Friday: first ship the beta, second update the docs, and third tell the team

Everything you'd fix by hand, already fixed.

Koi makes the smallest edits that turn speech into writing. When it isn't sure, it leaves your words alone.

Fillers and stutters, gone

Um, uh, you know — and the words you said twice.

Changes of mind, applied

Say “no wait” or “I mean” and Koi keeps only what you settled on.

Numbers that look like numbers

Times, prices, percentages and counts come out as digits.

Names spelled your way

Add names and jargon to your dictionary. While you dictate, Koi also checks the names and terms on your screen.

Learns from your fixes

Fix a name or term right after dictating, and Koi adds your spelling to your dictionary.

Any app, any input source

Works wherever there's a text field, whichever input source is selected.

Built to keep your words yours.

Koi goes online only to process what you dictate or edit by voice, and to learn a word you fix right after.

Only when you speak

When you dictate or edit by voice, Koi sends your recording, your dictionary, your name, and context from the app you're in: the text around your cursor, the window or page title, and words visible on screen. If you fix a word right after a dictation, that fix is sent too, so Koi can learn it.

Password fields are off-limits

Dictation and context reading switch off in secure input fields.

Falls back to the transcript

If the cleanup fails, times out or comes back broken, Koi types what it heard instead.

Your clipboard, put back

After pasting, Koi puts back what you had copied, as long as macOS lets it read the clipboard. Passwords copied from a password manager are cleared instead.

Questions

Which Macs does Koi run on?

macOS 26 or later, on a Mac with Apple silicon.

Do I need to switch input sources?

No. Koi works with whatever input source is selected. It pastes the text in for you, then puts your clipboard back. Pasting needs the Accessibility permission.

Which languages can I speak?

English, Chinese, Japanese and Spanish. English and Chinese can even be mixed in the same sentence. Dictate never translates: what you say in English stays in English.

Will it reword what I say?

Not in Dictate. Koi only removes fillers, applies your corrections, fixes misheard words and punctuation, writes numbers as digits, and lays the text out for the app you're in. Rewording is what Edit is for.

How long can I talk at once?

Up to 3 minutes per recording.

Easier said than typed.

Download for Mac

macOS 26 or later · Apple silicon · English, Chinese, Japanese and Spanish