AI in Rostrum is a capability of the platform, not a product you buy — and it follows one contract everywhere it appears: it proposes, you review, and nothing is written until you apply it.
AI assistance is enabled per organization by your administrator. When it is off, every AI surface in this chapter is simply absent — no grayed buttons, no upsell — and everything else in this guide works exactly as described. When it is on, three surfaces gain an assistant:
Reading your event website
The guided setup interview (Chapter 04) can read your event's public website and pre-fill the name, dates, venue and topic areas it finds — each shown for your confirmation, never silently applied. It reads only the page you give it.
Building the program conversationally
Build my program on the agenda opens a conversation. Describe what you want — "five rooms, an opening keynote on day one, lunch at noon every day, done by 11 on the last day" — and the assistant asks what it genuinely needs to know, then proposes a complete schedule: the rooms to create and every session with its title, day, times and room, laid out day by day for review. Follow-up messages revise the proposal; Create sessions applies it through the same validated scheduling machinery as the Session Builder, conflicts checked and all. You can also hand it a past year's program — its website, or the program text pasted in — and build from that.
Suggesting sessions for unplaced talks
With a tray full of unplaced presentations, Suggest sessions groups them by topic and proposes named sessions to hold them. The assistant reads titles and abstracts only — never author names or emails — and its groupings arrive as proposals you apply, adjust or ignore.
The rules it works under
- Review before write. Every AI proposal is shown to you in full before anything is created, and anything in a proposal that fails validation — an unknown room, an impossible time — is dropped and counted visibly rather than smoothed over.
- Your permissions, not its own. Applying a proposal uses the same permission-checked actions you would use by hand. The assistant can never do something you could not.
- Scheduling arithmetic stays deterministic. Where AI contributes, it contributes judgement — names, groupings, drafts; validation and conflict checking are code.
- No delegate-facing AI. Attendees, submitters and reviewers never interact with an AI surface.
- A monthly budget per organization caps spend; when it is reached, AI features rest until the next month and say so plainly.