CAMEL

CAMEL was the first to propose letting LLM agents autonomously cooperate through role-playing. The core method is "Inception Prompting" — using carefully designed role prompts to let two agents (e.g., "AI User" and "AI Assistant") autonomously converse to complete tasks.

The community has 100+ researchers. In 2024-2025, the OWL sub-project was accepted by NeurIPS 2025. The framework supports simulating up to 1 million agents and has fully integrated the MCP protocol. This pioneering research laid the foundation for later frameworks like AutoGen and MetaGPT.