Multi-agent
Handoff / swarm
Agents pass control to each other as the task shifts.
N agentsAdvanced
How it works
- 1Each agent has a narrow specialty and a set of peers it can hand off to.
- 2When the task moves outside its scope, it transfers control plus context.
- 3The receiving agent continues seamlessly.
- 4A shared context object travels with each handoff.
Use it when
Conversations that move between specialties mid-flight — sales hands to support, who hands to billing — without a central orchestrator.
Reach for something else when
You need tight central control or auditability. Decentralized handoffs are harder to trace.
Where you stay in the loop
Specialists pass control among themselves; you own the boundaries of who handles what. Decentralized handoffs are hard to trace, so keep a record you can follow afterward.
In the wild
A triage agent greets, then hands a technical question to a specialist agent, passing the full conversation along.
Hand this to your agent
Model a swarm with handoffs.
Define 3–4 specialist agents, each with a scope and the peers it can hand off
to. Start with the triage agent.
On each of my messages: the active agent responds, and if the topic leaves
its scope it announces a handoff ("→ handing to X because…") and the new
agent takes over, carrying the context. State which agent is speaking each
time.
Specialties: <...>Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.