A meeting is a live, round-based discussion between two or more teams and/or individual agents,
facilitated by the coordinator, that you can watch happen in real time and step into mid-conversation.
It's a different mode from delegation: instead of one team quietly doing work and reporting back,
several participants think out loud together, and you get a decisions list at the end you can turn
straight into new tasks.

From the Meetings section, start a new meeting: give it a topic, choose the participants
(any mix of whole teams or individual agents — each can run on a different model and provider, so
you can put a Claude-backed team in the same room as a free Ollama agent), and pick a **thinking
preset**:

| Preset | Rounds | Notes |
|---|---|---|
| Fast | 1 round | Quick take, no extended thinking |
| Balanced | 2 rounds | Default — enough back-and-forth to actually disagree and refine |
| Deep | 3 rounds | Longer turns, extended thinking enabled where supported |

Within each round, every participant speaks once, in turn: the coordinator emits a live "thinking →
speaking → done" status per participant, and each turn streams in as it's generated rather than
appearing all at once. Participants are explicitly instructed to react to what others already said,
not repeat themselves, and end with their single most important open question or recommendation.

Between rounds (except the last), the coordinator adds a short synthesis turn on its own lane —
points of agreement, real disagreements, and one or two focused questions the next round needs to
answer. It stays neutral here; it doesn't add its own proposals.

You can type a note while a meeting is running — it's queued and lands in the transcript before the
next participant's turn, visibly steering what gets discussed from that point on. This is the way to
narrow scope or redirect a meeting that's drifting, without stopping and restarting it.

A meeting ends automatically after its preset's round count, or you can click End meeting at any
point — it finishes the current turn, then cooperatively wraps up from whatever was said so far
rather than cutting off mid-thought.

When a meeting ends, the coordinator reads the full transcript and extracts a decisions list —
each one a concrete, actionable item, with a short rationale and (where one clearly fits) a
suggested owning team. From there you can dispatch any decision straight into a real background
task assigned to that team, without retyping anything.

The full transcript, summary, and decisions are saved back into the Brain under
Meetings/, tagged with the playground's Brain tags when the meeting was started from inside one —
so the outcome of a meeting is searchable later the same way any other note is.