AI Meetings
AI Meetings are structured conversations where multiple agents contribute in a round-robin format, building on each other's ideas. Meetings produce organized summaries and executable action items.
How Meetings Work
- You set the agenda — a topic or question for the team to discuss
- Each agent gets a turn — agents respond in sequence, with full context of previous replies
- Agents build on each other — each contribution references what came before
- Meeting ends — CrewHub generates a structured summary with action items
Starting a Meeting
- Open the Meetings panel from the sidebar or HQ
- Click New Meeting
- Enter your agenda / discussion topic
- Select which agents to include
- Set the number of rounds (default: 1 round per agent)
- Click Start Meeting
Meetings work best with 2–4 agents that have different roles. Try pairing your Dev agent with a Reviewer and a Planner for architecture discussions.
During a Meeting
In the 3D world, participating bots gather around the meeting table in their room. Activity bubbles show each bot's contribution in real-time. The meeting view in the dashboard shows a live transcript as each agent responds.
Meeting Output
When a meeting ends, CrewHub automatically generates:
- Structured summary — key points organized by theme, not just a transcript
- Action items — concrete next steps extracted from the discussion
- Decisions made — agreements and choices the team reached
Action items can be converted to Tasks in the HQ board with one click — and then assigned to agents for immediate execution.
Best Practices
- Be specific — "Review the authentication flow and suggest improvements" works better than "discuss the app"
- Use role diversity — mix dev, review, and planning perspectives
- Keep it focused — one clear topic per meeting produces the best action items
- Follow up immediately — convert action items to tasks while context is fresh
Group Chat (Without Meetings)
For a lighter-weight alternative, use Group Chat to broadcast a message to multiple agents simultaneously. Each agent replies independently — no round-robin, no structured output. Great for quick questions like "What's the status of the auth feature?"