Multi-Agent Collectives

What are Collectives?

Collectives are groups of agents that work together to solve complex problems. Each collective uses an orchestration strategy to coordinate how agents run, and an optional visual orchestration graph (nodes + edges) for strategies that need one. Collectives are managed through the REST API and the Python SDK — there is currently no dashboard page for them; the dashboard's node-based canvas for chaining agents/tools/conditions is Agent Studio (/workflows) instead.

Orchestration Strategies

The seven strategies returned by GET /api/v1/collectives/strategies:

• Sequential - Agents execute one after another, each building on the previous
• Parallel - All agents execute simultaneously, results aggregated
• Decision Graph - Routes based on a linked decision graph (decision_graph_id)
• Hierarchical - Supervisor agent delegates to worker agents
• Consensus - Multiple agents vote on decisions
• Router - Routes tasks to specialized agents based on content
• Custom - Design your own agent orchestration visually via the orchestration graph endpoints

Example: Create a Collective

POST /api/v1/collectives — request body:

{
  "name": "Research Team",
  "description": "Multi-agent research pipeline",
  "strategy": "hierarchical",
  "agents_config": [
    { "agent_id": "research-lead", "role": "manager", "order": 0 },
    { "agent_id": "data-analyst", "role": "specialist", "order": 1 },
    { "agent_id": "domain-expert", "role": "specialist", "order": 2 },
    { "agent_id": "fact-checker", "role": "specialist", "order": 3 }
  ]
}

agent_id values reference real agents created via POST /api/v1/agents (see the Agents API). Execute a collective with POST /api/v1/collectives/{collective_id}/execute.

Use Cases

• Research Teams - Multiple experts collaborate on analysis
• Content Review - Writer, editor, and fact-checker pipeline
• Code Review - Multiple reviewers with different focuses
• Decision Making - Agents debate pros and cons of options