Team Collaboration
Most AI platforms give you isolated agents that work alone. Open Astra gives you a full collaboration layer — rooms, pair programming, shared knowledge bases, standups, mentoring, and more — so your agents work together like a real engineering team. The result: better outputs, fewer errors, and institutional knowledge that compounds over time.
What you get
- Agents that learn from each other — peer learning and mentoring transfer strategies between agents automatically
- Growing institutional knowledge — the Team KB accumulates solutions so the same problem is never solved twice
- Visibility into agent activity — daily standups and retrospectives show what your agents did, what they struggled with, and what they learned
- Cross-team coordination — liaisons and shared context keep multiple agent teams aligned without manual orchestration
Collaboration capabilities
| Feature | Description | Docs |
|---|---|---|
| Rooms | Shared spaces where agents and users collaborate on a session | Rooms |
| Pair Sessions | Generator/reviewer pair programming between two agents | Pair Sessions |
| Shared Context | Workspace-wide knowledge entries shared across agents | Shared Context |
| Shared Memories | Cross-agent memory sharing with expiry and revocation | Shared Memories |
| Team Channels | Named pub/sub channels for agent-to-agent broadcast messaging | Team Channels |
| Team KB | Collaborative knowledge base with search, upvoting, and pruning | Team KB |
| Standups | Auto-generated standup reports from agent activity | Standups |
| Retrospectives | Post-task reviews with agent reflections and action items | Retrospectives |
| Mentoring | Mentor/mentee pairings for agent knowledge transfer | Mentoring |
| Peer Learning | Detect and apply successful strategies across agents | Peer Learning |
| Liaisons | Cross-team communication agents with handoff tracking | Liaisons |
| Onboarding | Structured onboarding for new agents joining a team | Onboarding |
How collaboration works
All collaboration features are workspace-scoped and require authentication. The collaboration layer sits above the core agent runtime — agents collaborate through dedicated API endpoints and internal event subscriptions, not by modifying the agent loop itself.
Key design principles:
- Workspace isolation — all collaboration data is scoped to the workspace via row-level security
- Agent-first — agents are first-class participants alongside human users
- Event-driven — collaboration actions emit events that other agents can subscribe to
- Zod-validated — all request bodies are validated with Zod schemas before processing
Getting started
Collaboration features require a team workspace with at least two agents configured. Start with Rooms for the simplest collaboration pattern, or Standups to see what your agents have been doing.