Swarm<T>
Models & Types · AiDotNet.Agentic.Agents
A peer-to-peer multi-agent team where control transfers between members over one shared conversation.
Unlike a SupervisorAgent (a hub that runs workers as subroutines), a swarm has no
central coordinator: whichever member is active answers directly, and may hand the whole conversation
to a peer, who then continues from the same history.
Implements: IAgent<T>
For Beginners
Imagine specialists passing a customer between them. The customer (the conversation) stays the same; whoever is currently helping can either answer or say “let me transfer you to my colleague”. This class runs that back-and-forth and gives you the final answer plus which member gave it.
How It Works
Handoffs are expressed as native tool calls (transfer_to_<peer>). The swarm intercepts these
at the loop level: rather than executing them, it switches the active member and re-runs the turn with
the new member’s instructions and tools against the unchanged conversation. A member’s own (non-handoff)
tools are executed normally. The whole team shares one MaxIterations budget,
which guarantees termination even if two members would otherwise ping-pong.
Constructors
| Constructor | Summary |
|---|---|
Swarm(IReadOnlyList<SwarmMember<T>> members, string entryMemberName, SwarmOptions options) |
Initializes a new swarm. |
Properties
| Property | Summary |
|---|---|
string Description { get; } |
|
string Name { get; } |
Methods
| Method | Summary |
|---|---|
Task<AgentRunResult> RunAsync(IReadOnlyList<ChatMessage> messages, CancellationToken cancellationToken) |
|
Task<AgentRunResult> RunAsync(string request, CancellationToken cancellationToken) |