Multi-Agent (Swarm)

Specialist routing

The Swarm pattern routes requests to specialist agents. The orchestrator LLM decides which specialist handles each request.

import { Swarm, Agent, anthropic } from 'agentfootprint';

const provider = anthropic('claude-sonnet-4-20250514');

const researcher = Agent.create({ provider, name: 'researcher' })
  .system('You are a research specialist. Be thorough and factual.')
  .maxIterations(3)
  .build();

const writer = Agent.create({ provider, name: 'writer' })
  .system('You are a writing specialist. Be creative and engaging.')
  .maxIterations(3)
  .build();

const swarm = Swarm.create({ provider, name: 'orchestrator' })
  .system('Route research questions to the researcher, writing tasks to the writer.')
  .specialist('researcher', 'Research topics and provide factual information', researcher)
  .specialist('writer', 'Write or rewrite content in a specific style', writer)
  .streaming(true)
  .maxIterations(5)
  .build();

const result = await swarm.run('Write a haiku about debugging');

How it works

The orchestrator sees each specialist as a tool. When it calls the researcher tool, the researcher agent runs as a subflow. The narrative captures the full execution path — orchestrator decision, specialist execution, and result.

Orchestrator → route to "writer" → Writer Agent runs → result returned → Orchestrator responds

Observability

swarm.getNarrative();
// [Seed] Initialized agent state
// [CallLLM] Orchestrator decided to route to writer
// [Subflow: writer] Writer specialist executed
//   [CallLLM] Writer generated haiku
// [Finalize] Final response assembled

Each specialist’s execution appears as a subflow — drill into it for full detail.