Kivo. Don’t type, just speak.

Dictation that stays on your Mac

Hold a key. Say it. Let go. Your words are there.

Download for Mac

Free to use. No account. No credit card.

A looping example. A person holds the Option key and says “um can you send me the the notes after the meeting”. Kivo types “Can you send me the notes after the meeting” into an email. One undo removes it.

You think in sentences.You speak in sentences.Then you sit down and type one letter at a time.

Kivo removes the letters. The rest is yours.

A small app. It does one thing.

One key. Hold it and speak. Let go and you are done. There is nothing else to learn.

Every app. Your words land at the cursor, wherever the cursor is, as one edit. One undo takes them back.

On only while you hold. The microphone opens when you press and closes when you let go. Your menu bar shows it.

No connection needed. Download the speech model once. Then turn off your Wi-Fi. Kivo works the same.

Nothing leaves. Audio is never uploaded and never saved. These are promises, not a live monitor.

Your words. Add the names and jargon dictation usually gets wrong. Kivo corrects them every time.

One voice. Six ways to write it down.

You say: “um the the report is ready”

An example. It does not use your microphone.

Three steps. That is all of them.

Hold.
Press your key in any app. The microphone opens.
Speak.
Say it the way you would say it. The “ums” and the repeats are taken out.
Let go.
The microphone closes. The speech model runs on your Mac and the text lands at your cursor.

Kivo v1, the speech model.

Speed, memory and load time were measured in August 2026 on Apple silicon.

Runs on: Your Mac, through Core ML. Your audio is not uploaded.
Download: 460 MB, once. It is checked against a pinned SHA-256 hash.
Offline: After the download, dictation needs no connection. You can remove the model in Settings.
Language: English only.
Output: Text with punctuation and capitalisation.
Speed: About 660 ms for 7.4 seconds of speech. The target is 300 ms.
Memory: About 960 MB while loaded. The target is 150 MB.
Load time: About 6.5 seconds when Kivo starts.

No account.No credit card.No upload.

Client files. Medical notes. Unreleased code. What you say stays on your Mac, because Kivo never sends it anywhere. The only thing that uses the network is downloading the speech model.

Straight answers.

Yes. It is free to use, with no account, no credit card and no paid tier.

That permission is how the global key is seen and how text reaches your cursor. Kivo reads nothing from your screen.

Yes. The model downloads once and is checked against a pinned SHA-256 hash. After that, dictation needs no connection.

No. Version 0.1.0 is a preview. It is unsigned, so macOS may warn you on first launch. It is English only and Mac only, and it is slower than its 300 ms target. Windows is planned.