Open Agent JSON

Visual, production-ready AI workflows — portable as JSON

Overview

Open Agent JSON enables you to define complex multi-agent systems and LLM-powered workflows using simple JSON configuration files instead of writing code. It provides a standardized format for:

• Multi-agent orchestration with LangGraph.js
• LLM model integration (OpenAI, Anthropic, Ollama)
• Agent-to-Agent (A2A) communication
• Model Context Protocol (MCP) server integration
• Interactive workflows with user interrupts
• Conditional routing and state management

Key Features

• Declarative Workflow Definition: Define nodes, edges, and state transitions in JSON
• Multi-Model Support: Seamlessly integrate OpenAI, Anthropic, and Ollama models
• A2A Protocol: Enable agent-to-agent communication for complex multi-agent systems
• MCP Integration: Connect external tools via Model Context Protocol servers
• State Management: Flexible state annotations with custom reducers
• Interactive Flows: Built-in support for user interrupts and approval gates

Quick Example

{
  "config": {
    "info": {
      "version": "1.0.0",
      "description": "Simple greeting workflow"
    }
  },
  "stateAnnotation": {
    "name": "GreetingState",
    "type": "Annotation.Root"
  },
  "annotation": {
    "messages": {
      "type": "string[]",
      "reducer": "(x, y) => x.concat(y)",
      "default": []
    }
  },
  "models": [],
  "nodes": [
    {
      "id": "greet",
      "handler": {
        "parameters": [
          {
            "name": "state",
            "parameterType": "state",
            "stateType": "typeof GreetingState.State"
          }
        ],
        "function": "const name = interrupt('What is your name?');\nreturn { messages: [`Hello ${name}!`] };"
      }
    }
  ],
  "edges": [
    { "from": "__start__", "to": "greet" },
    { "from": "greet", "to": "__end__" }
  ],
  "stateGraph": {
    "annotationRef": "GreetingState",
    "config": {
      "checkpointer": { "type": "MemorySaver" }
    }
  }
}

Documentation

For complete file format specification, see FILE_FORMAT.md.

The documentation covers:

• Complete schema reference for all sections
• Model configuration (OpenAI, Anthropic, Ollama)
• Node types (Function Nodes, Tool Nodes)
• Edge types (Normal, Conditional)
• State management patterns
• A2A and MCP integration
• Working examples and best practices

Examples

Explore real-world implementations:

• kudosflow: Example workflows including interactive flows, model integration, and A2A orchestration
• a2a-server: A2A server implementation for multi-agent systems

Technology Stack

This specification is powered by:

• LangGraph.js: Workflow orchestration framework
• SceneGraphManager v2.0.0: Private library for JSON-to-workflow compilation
• LangChain: LLM integration and tooling

License

This project is licensed under the MIT License - see the LICENSE file for details.

Note: This specification references the SceneGraphManager v2.0.0 library (kudos-scene-graph-manager), which is a private component. For licensing and access inquiries, please contact Akira Kudo.

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Author

Akira Kudo

• LinkedIn: akira-kudo-4b04163
• GitHub: @akudo7

Related Projects

• LangGraph.js - Official LangGraph.js documentation
• kudosflow - Example workflows
• a2a-server - A2A server implementation