Multi-Agent

Complex tasks often span several specialities. Multi-Agent lets one agent coordinate others — delegating sub-tasks to specialist agents and combining their results. You configure it from the Multi-Agent tab of the agent builder.

How it works

• A primary agent orchestrates the conversation.
• It can hand off to specialist agents, each with their own system instructions, knowledge and tools.
• Specialists work on their part and return results; the primary agent weaves them into a single coherent reply.

Think of it as a team: a coordinator plus focused experts, each strong at one thing, collaborating through coordinated handoffs and shared context.

When to use it

• A request naturally decomposes into distinct domains (e.g. billing + technical + policy).
• You want to reuse existing, well-tuned agents as building blocks rather than cramming everything into one giant prompt.

Cost & visibility

Inter-agent calls are metered like any other model usage and recorded as their own call type, so you can see exactly what the collaboration costs in the cost report (look for the interagent call type).

Tip: Keep each specialist narrow and well-instructed. Small, sharply-scoped agents are easier to test and compose than one over-loaded agent.

Next steps

• Testing & Conversations — exercise the whole team end to end.
• Publishing & Environments — promote the coordinated set together.