Meeting mode

Record a meeting, get a timestamped transcript, and have Luna write the minutes.

Record a meeting

1

Start

Click + in the message box and choose Meeting mode. The screen clears to a single orb.

2

Talk

The orb pulses with the audio and shows the elapsed time. Audio is uploaded in chunks as you go, so a long meeting isn't lost if the tab closes.

3

Stop

Click the orb, or press Stop. Luna transcribes the recording and writes the minutes, usually within seconds of a short meeting ending.

Meeting mode: a single glowing orb with a timer while recording
While recording there's nothing on screen but the orb.

What you get

  • A title suggested from what was discussed.
  • Summary of the meeting in a few sentences.
  • Discussion: the main points, topic by topic.
  • Decisions that were made.
  • Action items as a checklist, with an owner when one was named.
  • Open questions left for next time.
  • The full transcript with timestamps.
Meeting minutes with summary, decisions and action items
The minutes open in chat when processing finishes.

Where it's saved

Each meeting gets its own folder in the workspace and shows up in the Library:

workspace/meetings/<id>/
recording.webm    # the audio
transcript.md     # timestamped transcript
minutes.md        # the minutes
_meeting.json     # status, title, timings

How it works

  • Transcription runs locally with faster-whisper (the same model as voice, set by stt_model). Nothing leaves your server unless you chose a cloud model.
  • Long transcripts are summarized in parts, then combined, so they fit any model's context window.
  • The minutes are written by whichever AI model you've chosen in Settings.

API

MethodPathDoes
POST/api/meetingsStart a meeting; returns its id
PUT/api/meetings/{id}/audio?seq=NUpload audio chunk N
POST/api/meetings/{id}/finishStop and start processing
GET/api/meetings/{id}Status, transcript and minutes
GET/api/meetingsList meetings

Limits

📌
No speaker labels yet

The transcript doesn't say who said what, so action-item owners are only filled in when someone is named in the conversation.

Tell people they're being recorded before you start.