Skip to main content

How dictation works

Hold fn anywhere on your Mac and speak. Everything — speech recognition and cleanup — runs on-device.

J
Written by Junan Pang

Amlin's dictation is system-wide: it works in any text field in any app — Slack, Mail, Notion, VS Code, Cursor, a browser form, an Amlin note.

The interaction

  1. Put your cursor where you want the text.

  2. Hold fn and speak.

  3. Release. Clean text appears at the cursor.

A small pill shows up while you're speaking, with a live waveform, elapsed time, and a cancel affordance. It never steals focus from the app you're typing into.

It runs entirely on your Mac

  • Speech recognition runs on-device (a local model downloaded once, ~450 MB).

  • The cleanup that removes filler and fixes punctuation also runs on-device.

  • No audio and no text leaves your Mac. There is no cloud dictation mode in this version.

  • Dictation works on a plane, on hotel wifi, or with the network off entirely, and it costs nothing against your plan — dictation is unlimited on every plan, forever.

What "cleanup" actually does

Say: "so um I think we should, we should push the launch to Wednesday, actually Tuesday, and then first we need the copy second the screenshots third legal sign off"

Get: "I think we should push the launch to Tuesday, and then we need: the copy / the screenshots / legal sign-off" — filler removed, the self-correction resolved, the enumeration turned into a list, punctuation applied.

The safety rules

  • It never presses Return for you. Dictated text can never submit a message or a form — any newline in the output is flattened to a space.

  • It refuses secure fields. Password fields are detected and skipped entirely.

  • It never loses your words. If insertion fails for any reason, the text goes to your clipboard with a notification, and it's always in History.

  • It inserts where you started. If you switch apps mid-dictation, Amlin won't spray text into whatever now has focus.

Did this answer your question?