Talk to Roma with your voice

You can say anything to Roma instead of typing it. In the browser and in the Mac app, tap the mic in the composer, or hold a key anywhere on your Mac and talk. On the iPhone, tap the mic, or press the Action Button. Your words are transcribed and handed to the operator, which turns them into tasks, notes or an answer exactly as it does with typed words.

In the browser and in the Mac app

The composer shows a mic button while it is empty. Tap it and recording starts at once: the button becomes a pill with an X, a live waveform and a stop square, and your words appear as you speak. The stop square puts the transcript into the composer so you can read it over, and Enter sends it. Escape cancels the take, and so does typing into the field. The X cancels the take and clears the draft, keeping any attachments. A take is at most two minutes long. The same mic sits on every chat composer, on your main conversation and on an item's.

Hold to talk on the Mac

On the Mac you can talk to Roma from any app without opening Roma first. Open Settings, choose Shortcuts and switch on Enable hold to talk, then pick a trigger: the Globe key, Shift with Globe, or Right Command are offered, and you can record any other key. macOS asks you to grant Accessibility once, because Roma has to watch for the key.

Then hold the trigger anywhere, in any app. The quick chat panel appears without taking your keyboard away from what you were doing, already listening, and your words land in it as you speak. Release the key and the take parks in the panel: press Enter to send it, or Escape to discard it. Switch on Send on release and releasing the key sends straight to your operator, the panel slips away, and the operator gets to work in the background while you stay where you are. The quick chat hotkey (Control, Option and N by default) opens the same panel idle, for typing or pasting a screenshot. The panel is covered in quick chat and hold to talk.

On the iPhone

Tap the mic on the chat pill and talk for as long as you like: a one-line task or a ten-minute think out loud, and nothing is ever cut short. The pill shows an X, a waveform and an arrow. Tap the arrow, or anywhere on the pill, to stop and send in one go. Tap the X to cancel the take and clear the draft. The mic is also on the keyboard bar wherever you write, so you can dictate into the New task row or into a note's body, and the words land in the field, where you edit them like typed text.

One setting decides when spoken words appear, on every mic in the app. Open Settings, choose AI and Dictation, and under Dictation pick Live or After recording. After recording is the default: nothing appears while you talk, and the whole take is transcribed once you stop, which is the more accurate of the two. Live shows the words as you say them, less accurately. A take over about half a minute is transcribed in pieces as you go, so even a long one lands quickly when you stop.

The Action Button opens Roma listening, in your main conversation, and the mic is live by the time the app is on screen. Under Settings and Action Button you choose Mode, Voice or Text, and in voice mode two more switches: Send after silence ends the take after three seconds of quiet and sends it after a short countdown you can cancel, and Talk without opening Roma lets you talk from the Dynamic Island without opening the app at all: you press the Action Button again to send, and the operator's answer arrives as a notification. All of that is in the Action Button, and the rest of dictation on the phone is in voice and dictation on the iPhone.

What happens to the words

A spoken message reaches the operator as the words you said, and a question gets an answer while the rest becomes a task, a note or an item, as what the operator does with what you say explains. A capture you dictate is tidied in the background, with a clean title and the due date read out of the words, to the Level and Voice you chose under AI editing. Saying an expert's name calls that expert in, with your words left exactly as you said them. The operator replies in your language.

Sam is reading a long email in the browser, holds the Globe key and says "remind me to send Noor the contract before lunch and book the dentist for next week", then lets go. With Send on release switched on, the panel slips away and Roma has two tasks a moment later: Send Noor the contract, due today, and Book the dentist, due next week. Sam never left the email.

Questions people ask

Do I have to stop talking after a few seconds?

No. On the iPhone a take has no limit, and a long take is transcribed in pieces as you go. In the browser and in the Mac app a take runs up to two minutes, which is a few paragraphs of speech.

Can I read the words before they are sent?

In the browser and in the Mac app, yes: the stop square and the hold-to-talk release both park the transcript in the composer, and Enter sends it. On the iPhone the arrow on the chat pill sends as soon as it stops. Dictating into a field, such as the New task row or a note, lands the words in the field, where you can edit them before you save.

Why do the words only appear once I stop talking on the iPhone?

Because Dictation is set to After recording, which is the default. The whole take is transcribed in one pass once you stop, which is more accurate than transcribing live. Switch to Live under Settings and AI and Dictation if you would rather watch the words land as you speak.

Does Roma understand my language?

The operator replies in the language you speak to it, and whatever it saves stays in the language you said it in. A dictated capture is tidied in that language and never translated.