Meetings

Record a meeting, read its transcript and ask PHNTM to draft a summary on this device.

How to

Start and stop

  1. Tell everyone before recording and get any consent required where you are. Read and accept the recording notice before starting.
  2. Open Meetings from Home or Notes. Choose the meeting language; it starts with the device language. English, Italian and Spanish use the local recognition route.
  3. Put the device near the speakers. As a starting recommendation, keep people within about 2 metres and have one person speak at a time. This distance is advice, not a measured accuracy limit.
  4. Tap Start recording. The recording indicator and elapsed time stay visible; the meeting view shows the transcript. Each piece of text has a minutes:seconds time mark. Starting from a paired phone also shows the indicator on the device.
  5. Tap Stop recording when finished. Recording ends, but queued audio may still be turning into text. Wait for that work to finish before reviewing the complete transcript.

Review and use the text

  1. Open the saved meeting to read or search its transcript and rename it.
  2. For another recognition pass, tap Improve accuracy. Progress is shown and you can cancel. The existing transcript stays until the new pass finishes. When it finishes, the new text replaces the transcript and any earlier summary is cleared; tap Summarise again afterwards if you need one.
  3. Tap Summarise for a draft labelled Summary written by PHNTM from the transcript. Review it against the transcript: it can omit or misstate things. A partial result is not a finished summary.
  4. Use Save to Notes to make a note (long transcripts become numbered notes), or Send to my phone/computer to download text or Markdown (.txt or .md) through a paired device.
  5. Use Delete audio, keep transcript to free the recording space while keeping the text. Without that audio, you cannot run the accuracy-improvement pass. Delete meeting removes the meeting from this device, including its audio and transcript.

Good to know

If something goes wrong

What you seeWhat to do
The meeting speech model is not installed or cannot be verifiedRecording is unavailable until the required model is installed correctly. Contact support; there is no silent switch to a smaller model.
Text is catching up after StopLeave the device powered on to finish the queued audio. You can stop recording even if recognition is stuck.
The microphone disconnects or storage runs lowRecording stops and the saved part is kept. Reconnect the microphone or free space, then review what was saved.
A meeting was interrupted by a restart or power lossOpen Meetings and choose the offered recovery. Saved segments can be transcribed; the segment being recorded at the interruption may be missing.
A summary cannot start because the model is busy or unavailableYour transcript is still saved. Try again when the local model is available; the summary is not sent to a cloud service.

Privacy

Recording, recognition and meeting summaries run on this device, including in Boosted mode. Meeting audio is kept until you delete it. Meeting audio is not included in backups; transcripts are. A meeting is not automatically added to Chat or assistant memory. A copy saved to Notes follows the Notes controls: if you enable it for Chat, selected text can be used in answers and sent to your configured cloud or computer provider. A download goes to the paired phone or computer you choose. Take Everything does not include Meetings; to keep a copy, use Send to my phone/computer or Save to Notes.

Deleting a meeting removes its audio and transcript from this device. Copies already saved to Notes, downloaded, exported or included in earlier backups are separate; delete those separately if needed. Recording consent is your responsibility; tell everyone before you start.

More detail

The measurements available on 17 November 2026 used one far-field English meeting. The full 29.7-minute test used an English-only recognition model; each 20–30-second block took on average 23 seconds to transcribe, and up to 152 seconds while chat was answering. The delay from speech to on-screen text was not measured. These results do not establish Italian or Spanish recognition quality. The larger-model pass took 369 seconds for a five-minute sample, about 1.2 times the audio length, measured with a faster setting than the device uses; on the device, expect it to take longer than the recording. A full-meeting pass was not measured. Processing time and recognition quality vary with the recording and other work on the device.

Related topics

From the device manual for software 0.9.0 (November 18, 2026). Your device serves the manual for its installed software; where this copy and your device differ, your device is right.