Graph Compose

Multi-Agent Workflows 👥

🚀 Beta Feature!

Unlock advanced automation by orchestrating multiple AI agents that collaborate within a single workflow. Learn how to design, connect, and manage systems where agents specialize in different tasks.

Multi-agent workflows allow you to:

  • Break complex tasks into specialized agent roles
  • Share results between agents using dependencies
  • Reuse tools across different agents
  • Create sophisticated processing pipelines

How Agents Work Together

In Graph Compose, agents collaborate through:

  1. Dependencies - Agents can depend on results from other agents
  2. Shared Tools - Multiple agents can use the same tools
  3. Result Passing - Results flow through the workflow graph

Here's a visualization of a multi-agent workflow:

flowchart LR A[Research Agent] -->|Research Results| B[Writing Agent] B -->|Draft Content| C[Review Agent] D[Knowledge Base] -->|Used by| A D -->|Used by| B D -->|Used by| C style A fill:#4F46E5,stroke:#4338CA,color:#fff,rx:10 style B fill:#4F46E5,stroke:#4338CA,color:#fff,rx:10 style C fill:#4F46E5,stroke:#4338CA,color:#fff,rx:10 style D fill:#10B981,stroke:#059669,color:#fff,rx:10

Building Multi-Agent Workflows

Here's how to structure a workflow with multiple agents:

{
  "nodes": [
    {
      "id": "research_agent",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/research",
        "method": "POST"
      },
      "tools": ["kb_search", "web_search"]  // Tools for research
    },
    {
      "id": "writing_agent",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/write",
        "method": "POST"
      },
      "dependencies": ["research_agent"],    // Uses research results
      "tools": ["kb_search", "content_gen"]  // Tools for writing
    },
    {
      "id": "review_agent",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/review",
        "method": "POST"
      },
      "dependencies": ["writing_agent"],     // Reviews the written content
      "tools": ["kb_search", "analyze"]      // Tools for review
    }
  ],
  "tools": [
    {
      "id": "kb_search",                     // Shared knowledge base tool
      "type": "http",
      "http": {
        "method": "GET",
        "url": "https://api.example.com/kb/search"
      }
    },
    {
      "id": "web_search",                    // Research-specific tool
      "type": "http",
      "http": {
        "method": "GET",
        "url": "https://api.example.com/web/search"
      }
    },
    {
      "id": "content_gen",                   // Writing-specific tool
      "type": "http",
      "http": {
        "method": "POST",
        "url": "https://api.example.com/generate"
      }
    },
    {
      "id": "analyze",                       // Review-specific tool
      "type": "http",
      "http": {
        "method": "POST",
        "url": "https://api.example.com/analyze"
      }
    }
  ]
}

Common Patterns

Content Creation Pipeline

This pattern uses multiple agents to create and refine content:

{
  "nodes": [
    {
      "id": "researcher",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/research",
        "method": "POST",
        "body": {
          "task": "Research AI trends",
          "depth": "comprehensive"
        }
      },
      "tools": ["web_search", "kb_search"]
    },
    {
      "id": "writer",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/write",
        "method": "POST",
        "body": {
          "style": "technical",
          "format": "blog"
        }
      },
      "dependencies": ["researcher"],
      "tools": ["content_gen", "kb_search"]
    },
    {
      "id": "editor",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/edit",
        "method": "POST",
        "body": {
          "check": ["grammar", "style", "accuracy"]
        }
      },
      "dependencies": ["writer"],
      "tools": ["analyze", "kb_search"]
    }
  ]
}

Data Analysis Pipeline

This pattern combines agents for data processing and analysis:

{
  "nodes": [
    {
      "id": "data_collector",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/collect",
        "method": "POST",
        "body": {
          "sources": ["market_data", "social_media"],
          "timeframe": "last_week"
        }
      },
      "tools": ["data_fetch", "clean_data"]
    },
    {
      "id": "analyzer",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/analyze",
        "method": "POST",
        "body": {
          "analysis_type": ["trends", "patterns"]
        }
      },
      "dependencies": ["data_collector"],
      "tools": ["analyze", "visualize"]
    },
    {
      "id": "reporter",
      "type": "agent",
      "http": {
        "url": "https://api.example.com/agents/report",
        "method": "POST",
        "body": {
          "format": "executive_summary"
        }
      },
      "dependencies": ["analyzer"],
      "tools": ["summarize", "format"]
    }
  ]
}