Kudu docs

Kudu is voice dictation that runs entirely on your computer. Hold a shortcut, speak, let go: your words appear at the cursor, cleaned up and formatted, in any app.

How Kudu works

1. You hold the shortcut and speak. The microphone opens only while you dictate. A small pill near the bottom of the screen shows that Kudu is listening.
2. Speech becomes text on your computer. An open speech model (NVIDIA Parakeet) transcribes what you said: on the Apple Neural Engine on a Mac, on the CPU on Windows and Linux. Long dictations are transcribed in pieces while you speak.
3. The text is cleaned up. Fillers like “um” go, corrections like “Tuesday, no, Wednesday” are applied, “new line” becomes a line break, times and numbers are written properly, and names are spelled the way your dictionary has them.
4. Optionally, it’s formatted. Lists become lists and emails get a greeting and a sign-off, using Apple Intelligence on a Mac or a small downloadable model on Windows and Linux.
5. It’s typed where your cursor is, in the style you chose for that kind of app.

Where Kudu runs

Kudu understands English today. It’s open source under the MIT license; the code is on GitHub.