AI Adviser
The AI Adviser is an agent that sits beside a brainstorm and shares its perspective on the discussion — in a side panel, not in the thread. It reads the conversation up to the latest message and tells you where it thinks things stand: what's strong, what's missing, what to weigh next. Nothing it says lands in the thread unless you choose to bring it in — so it's a low-friction second opinion that doesn't interrupt the flow, and no one has to visibly "invite a bot" into a shared discussion.
Turning it on
Set an adviser when you create or edit a brainstorm: open its settings, switch AI Adviser on, and pick the agent that should advise. It's off by default — off means no background agent advises the brainstorm, and you can still call agents directly in the discussion with an @mention.
The panel
A lamp button in the brainstorm's top bar opens the Adviser panel. It shows the current advice — the adviser's take on the discussion up to the latest message — with a short preview of the message it's responding to (click the preview to jump to that message in the thread).
- Bring into thread — post an advice (or a selected part of it) into the shared brainstorm as a normal message. This is what makes an adviser point public: your call, not the agent's.
- History — every advice is kept. Step back through previous versions to see how the adviser's take evolved as the conversation developed. You can review them at any time; when a new advice arrives, the panel jumps back to the latest.
- Try again — if an advice fails, retry it.
When it runs — and why that saves tokens
The adviser is lazy on purpose: it does nothing while the panel is closed for everyone. If no one is watching, nothing is computed — that's the token-saving trick.
- As soon as any participant opens the panel — or the panel is already open when a new message arrives — the adviser runs on the latest message.
- A new message always wins. If someone posts while the adviser is still thinking on an earlier message, that in-flight run is cancelled and a fresh one starts on the newest message — so the advice never reflects a stale, half-finished view of the discussion, and no tokens are spent completing a take that's already out of date.
- The run is shared: one request regardless of how many participants have the panel open, and everyone sees the same advice and the same live "thinking → done".
- Re-opening the panel shows the existing advice (and its history) — it doesn't re-run needlessly.
Reading the button
The lamp button doubles as a status light, the same for everyone:
- Outline — no advice yet for the latest message.
- Pulsing (outline ↔ filled) — the adviser is thinking.
- Filled — there's a ready advice for the latest message.