In a browser
Every composer, in Sensei and in a chat session, has a microphone button, Dictate.
- Press Dictate. The first time, your browser asks for the microphone.
- Speak. The composer shows a red dot, the time and a sound meter.
- Press the check, Stop and transcribe, to finish. Press the cross, Discard recording, to throw the take away.
While Sensei writes it down, the composer shows Transcribing, with a cross to cancel. Your words go where the caret is, with a space against the words around them, so you can read and change them before you send. Nothing is sent by itself.
If you blocked the microphone, the button reads Microphone blocked: allow it in your browser's site settings. A browser that cannot record shows no Dictate button.
In a narrow window, Sensei uses the web's phone layout, where you hold New thread to talk, or press Talk on a card. See Deck.
On a Mac
In the desktop app, hold fn to dictate. On newer keyboards, it is the key with the globe.
- In a composer. Put the caret in it, hold fn and speak. The composer reads Release fn while it records. Let go and your words go into the composer, as with Dictate. They are not sent.
- In Spotlight. Hold fn to dictate into the Spotlight search box. Hiding Spotlight always turns the microphone off. See Spotlight.
- In Deck. Hold fn to record a reply to the card on screen, and let go to send it. With the card's reply already open, holding fn dictates into it instead. Press ? in Deck for the shortcuts.
A press of fn shorter than 0.3 seconds is a tap, and records nothing, so macOS keeps its own fn action. Pressing another key while you hold fn, such as an arrow, cancels the take. Dictation records only into the Sensei window in front. It never types into another app.
The desktop app needs the Microphone permission to hear you. See Permissions. To record calls, see Record calls.
On iPhone
In the iOS app, hold to talk and let go to send. Every hold works the same way:
- Hold still for a moment. The control turns into a red microphone, and a glass recording bar shows a red dot, the time, a sound wave and "‹ Slide to cancel".
- Let go to send.
- Slide sideways to throw it away. The microphone becomes a bin, and the bar reads "Let go to discard". Slide back to keep recording.
- Slide up to lock it, hands free. The bar then shows the bin at its start, a small lock and Send at its end.
- A short tap does what a tap does, such as opening the composer.
The first time, iOS asks for the microphone, and that take goes on hands free once you allow it. If the microphone is not allowed, the composer opens and says to allow it in iOS Settings.
New thread. Hold the Sensei button in the tab bar. The tab bar's glass becomes the recording bar. When you let go, your words go as a new thread at once, written down in the background.
Reply to a card. On Home, hold Reply on the card. The card's buttons become the recording bar. When you let go, the card goes, and Undo holds your reply for 5 seconds before it is sent. Undo drops it, words and all. If nothing could be heard, the card comes back with "The reply was not sent." and the reason. A spoken reply never changes a card's drafted reply: it is a reply to Sensei, as Reply is. A tap on Reply still opens the composer.
In a thread. The microphone in a thread's composer is hold only. A tap shows "Hold to talk". Let go and the message shows in the thread at once, with its recording to play and its words under it. While Sensei works, a hold steers it by voice.
Dictate into the composer. In the new-thread composer, press the microphone. The field reads "Listening" while the recording bar fills the row. Press the check to stop: the recording goes in the composer's tray, where you can play it, and its words replace it in the field once they are written down. The bin throws the take away and leaves what you typed as it was.
A voice message that could not be sent says why, and keeps its recording so you can try again.
On Apple Watch
The Watch app is not in the current iOS app yet. On the Watch, press the microphone on a card to reply, or open + New thread, which starts listening at once. Send stops, writes it down and sends in one step. See Reply and start a thread by voice.
On glasses
On Meta Ray-Ban Display glasses, you talk to Sensei through Meta AI. Ask it, for example, what needs you in Sensei. Meta AI can:
- List your threads that need you, up to 8 aloud, each with its name and why it needs you.
- Reply in one of those threads, named as you say it, and read you Sensei's answer.
- Answer the question on a thread's card, by the option's number, its title or your own words.
- Start a new thread with your words.
Meta AI waits up to 60 seconds for Sensei's answer; a later one stays in the thread on the glasses.
On the glasses' own screen, Reply opens a composer, and the glasses offer their own dictation for it. When Sensei's message is the last in a thread, Read aloud reads it through the glasses' speakers, and Stop ends the reading. Reading aloud is in English only, with one voice.
Where your voice goes
- A browser and the Mac. The recording stays on your device until you stop. Then it goes to Sensei once, to be written down, and only the words come back into your composer.
- iPhone. The recording is written down in the background, then your words are sent. A recording that has not gone yet stays on the phone until it is sent or you sign out. In a thread, the recording stays with its message to play while the app runs.
- Glasses. Meta AI and the glasses' own dictation hear you, and only the words reach Sensei.
Sensei writes recordings down with Google Gemini, in the language you speak, with no language to choose. Sensei's servers keep no copy of the recording. Writing it down does not use your AI provider.
Limits
- A recording stops by itself at 5 minutes. In a composer, it is then written down as if you had stopped it. On iPhone, a held new thread waits in the composer unsent, and a card reply or a thread message is sent.
- A recording can be up to 10 MB.
- Each hour, you can send up to 60 recordings and 60 minutes of audio. A workspace's composers together can send up to 600 recordings and 10 hours. Past that, Sensei says "You've reached this hour's dictation limit" and when to try again.
- Two of your recordings can be written down at once. A third is refused: "Another recording is still being transcribed. Try again in a moment".
- A recording with no words in it is not sent: "Nothing was heard in that recording."
Next steps
- Cards and the Deck for what a reply does to a card.
- Replies in your voice for how Sensei drafts the messages it sends as you.
- Mac to set up the desktop app and its permissions.
- Glasses to set up Sensei on Meta Ray-Ban Display.