AgentCreator — An AI Agent That Creates… More AI Agents 🤖➡️🤖

AgentCreator is an autonomous AI system that generates new AI agents using dynamic Python templates — inspired by the iconic self-replicating robot sequence in ROBO (Enthiran).

Built on Microsoft Autogen, it forms an evolving mini-ecosystem where agents collaborate, refine ideas, and create new agents with unique personalities and goals.

---

🚀 What AgentCreator Does

AgentCreator acts as a factory agent that:

• Reads a base template (agent.py)

• Generates entirely new agents with:

  • Custom system messages
  • Unique objectives
  • Different sectors/interests

• Writes the Python file (e.g., agent5.py)

• Dynamically loads & registers the new agent

• Sends messages between agents to refine ideas

• Produces 10+ AI-generated business ideas, each from a different agent mindset

The result is a network of interacting agents capable of creativity, iteration, and self-expansion.

---

🧩 Architecture Overview

1. Creator Agent (creator.py)

The mastermind. It:

• Reads the agent template
• Prompts an AI model to modify and generate a new version
• Saves the new agent file
• Dynamically registers the new agent with the runtime
• Requests an idea from the freshly created agent

2. Base Agent (agent.py)

Each generated agent:

• Has a defined personality & domain interests
• Creates or refines business ideas
• May “bounce” ideas to other agents for refinement
• Uses Autogen’s AssistantAgent under the hood

3. Messaging Layer (messages.py)

Defines a simple message datatype. Also chooses random recipient agents for cross-agent refinement.

4. Runtime Layer (world.py)

The orchestrator.

• Spins up the gRPC Autogen runtime
• Registers the Creator agent
• Creates multiple agents (agent1.py to agent10.py)
• Stores each agent’s idea into separate markdown files

---

🔧 Tech Stack

• Microsoft Autogen Core
• Autogen AgentChat
• OpenAIChatCompletionClient
• Python 3.10+
• gRPC Worker Runtime
• Dynamic module creation
• Auto-registration of agents

---

📦 Project Flow

Creator Agent
     │
     ├── Generates Agent Code (agent1.py … agent10.py)
     │
     ├── Registers Agent with Runtime
     │
     ├── Sends "Give me an idea"
     │
     └── Writes output → ideaX.md

Each generated agent has its own worldview — healthcare, education, fintech, entertainment, etc.

---

🏁 How to Run

Clone the repo:

git clone https://github.com/Anand-Velpuri/Agent-Creator.git
cd AgentCreator

Install dependencies:

uv sync

Set up environment variables:

BASE_URL=...
API_KEY=...
MODEL_NAME=...

Run the world:

uv run world.py

This will:

• Start the Autogen runtime
• Create 10 new agents
• Generate 10 unique business ideas
• Save outputs as idea1.md to idea10.md

---

🧠 Why This Matters

AgentCreator demonstrates the next evolution of AI systems:

AI that builds AI. AI that collaborates with itself. AI that scales creativity.

It’s a small step toward autonomous multi-agent ecosystems capable of innovation, coordination, and self-improvement.

🔗 Connect

Feel free to open issues, contribute, or discuss ideas. This space is evolving fast — and we’re just getting started.