AutoGen Creator World

Ed Donner

An AutoGen Core runtime where a Creator agent invents new worker agents from a template, then those workers generate business ideas — optionally bouncing ideas off each other.

AutoGen Core · AutoGen AgentChat · gRPC runtime · OpenAI

Overview

AutoGen Creator World runs a local gRPC AutoGen host (5_autogen/world.py). A Creator agent writes new Python agent modules from a template; Worker Agents then produce idea markdown, sometimes bouncing ideas to peers.

System design

  • Runtime host first. GrpcWorkerAgentRuntimeHost starts; workers register against it.
  • Create then message. Creator materializes agentN.py, registers the agent, then asks it for an idea.
  • Batch world loop. world.py fans out many create-and-message coroutines and writes ideaN.md files.

Product path

What a developer can run. Diagram key: ✦ AI agent · purple = actor · white = service.

---
config:
  theme: default
  themeVariables:
    background: "#ffffff"
    fontSize: "14px"
---
flowchart TB
    person[Developer]
    host[gRPC runtime host]
    creator["✦ Creator"]
    workers["✦ Worker Agents"]
    ideas[idea markdown files]

    person -->|start world.py| host
    host --> creator
    creator -->|new Agent modules| workers
    workers --> ideas --> person

    classDef agent fill:#e8f0ee,stroke:#0f3d36,stroke-width:2px,color:#0d0d0d
    classDef service fill:#fafaf8,stroke:#c8c8c2,stroke-width:1px,color:#0d0d0d
    classDef actor fill:#f3eef8,stroke:#6b4f9a,stroke-width:2px,color:#0d0d0d
    class creator,workers agent
    class host,ideas service
    class person actor

Layers and Anthropic names live on Mental Model and the agent tabs.

Clean product overview and path. Mental Model holds the layer map and Anthropic workflows; Creator and Worker Agents own role detail.