Meetings and voice
Two-track meeting recording, on-device transcription and speakers that are remembered.
Voice notes
⌘⇧V (Win+Alt+Shift+V on Windows, Super+Shift+V on Linux) starts a voice note, and the same keys again stop it. It is transcribed on your computer and the text lands on your clipboard, ready to paste, with the rest of what the app gathered. Pick the language in Settings, Transcription.
The app keeps more than the words: screenshots when the screen changes, the app and page at each moment, and what you clicked, in a context pack. Meetings do the same, so a screen share reads stage by stage. This is always on. To stop screenshots for one recording, click the screen icon on its panel; what was saved is dropped.
Meetings
⌘⇧M (Win+Alt+Shift+M, Super+Shift+M) starts recording a meeting and stops it. It records two tracks at once:
- Your microphone: you, and anyone in the room with you.
- The call audio: everyone on the other end, straight from your computer’s sound output. No bot joins the call.
On macOS and Windows the call audio comes straight from the system’s sound output. On Linux it comes from the output’s monitor in PipeWire or PulseAudio, where your system has one; without it, a meeting records your side only.
When a call app such as Zoom or Teams (or FaceTime on a Mac), or a call in your browser such as Google Meet, starts using the microphone, a small card asks once whether to record it. This works on macOS and Windows, and on Linux when PipeWire’s or PulseAudio’s tools (pw-dump or pactl) are installed. It won’t ask during your own recordings. Decline the card to skip a call. To start recording calls without the card, turn on Start recording calls without asking in Settings, Capture. Recordings under two seconds, or short and silent, are thrown away as accidents.
You are responsible for telling people they are being recorded, and for getting the consent the law asks for where you are.
Transcription
Transcription runs on your computer. On macOS 26 and later it uses Apple’s own transcriber; on Windows, Linux and earlier macOS it uses Whisper, with a speech model the installer downloads (or that you choose in Settings). Audio never leaves your computer and there is no usage limit. In packs, transcripts are tidied: filler words such as “um” are removed and the raw text is kept in manifest.json.
Who said what
After a recording, an on-device model separates the voices. It is built in on macOS; on Windows and Linux add the optional Speaker names download, in setup or Settings. Each voice gets a lasting identity, so naming “Voice 3” as Priya once means Priya is recognised in later recordings. A voice needs about eight seconds of speech before it becomes a lasting identity; shorter fragments are treated as noise.
Your own voice is learnt from the microphone track. Once it has turned up in a few recordings, it is labelled Me. Speaker recognition can be turned off in Settings.
**Me** · 00:12 Can we ship the new checkout behind the flag first? **Priya** · 00:19 Yes. I'd rather watch the decline rate for a week. **Voice 3** · 00:31 I can pull the numbers from Stripe on Friday.
After the call
The meeting’s group becomes a context pack and everything in it is copied to your clipboard as text. With an AI key set, it can also get a title and a summary.