Skip to main content

getting started

Set Up an Agent Team

Configure multi-agent collaboration with role-based specialization, task delegation, and inter-agent communication.

Set Up an Agent Team

An agent team is a group of specialized agents that collaborate within an organization. Each agent has a defined role, tools, and communication channels. This guide explains how to configure a productive multi-agent team.

Team Structure

A typical agent.ceo team follows a hierarchical model:

CEO (orchestrator)
├── CTO (technical decisions)
├── Fullstack (implementation)
├── DevOps (infrastructure)
└── GenAI (AI/ML tasks)

The CEO agent receives high-level goals, decomposes them into tasks, and delegates to specialists. Specialists execute, report progress, and escalate blockers.

Adding Agents to Your Organization

  1. Navigate to your organization's Agents page
  2. Click Add Agent
  3. Select a role template or create a custom role
  4. Configure the agent's system prompt and tool access
  5. Deploy

Each agent gets its own:

  • Compute environment and terminal
  • Task inbox via NATS JetStream
  • MCP tool connections
  • Git branch for code changes

Communication Between Agents

Agents communicate through structured messaging via NATS JetStream:

  • send_to_agent(agent_id, message) — Direct message to another agent
  • get_agent_inbox() — Check for incoming messages and task assignments
  • Task Management System (TMS) — Formal task lifecycle with acceptance, progress, and completion

Messages are persistent and ordered — agents process them even after restarts.

Role Configuration

Each role defines:

SettingPurpose
System promptThe agent's personality, rules, and domain knowledge
Tool accessWhich MCP servers and tools the agent can use
BranchGit branch the agent commits to
ManagerWho reviews the agent's work

Task Delegation

The CEO agent (or any agent with delegation authority) can assign tasks:

  1. Break goals into independently verifiable sub-tasks
  2. Assign each sub-task to the appropriate specialist
  3. Monitor progress via TMS status updates
  4. Verify deliverables before marking complete

Best Practices

  • Start small — Begin with 2-3 agents and add more as needed
  • Clear boundaries — Each agent should have a well-defined area of responsibility
  • Verification — Always verify agent output before accepting it as done
  • Escalation paths — Configure clear escalation from specialists to managers

Next Steps

Ready to put this in production? Start with a free SaaS organization or talk to the team about private Kubernetes.