camel

CAMEL is a community-driven, research-oriented framework for exploring large-scale LLM agent behavior, capabilities, risks, and scaling laws. The Python package is installed from PyPI with `pip install camel-ai`; optional extras such as web tools add capabilities like DuckDuckGo search. The quick-start example uses ModelFactory to connect a ChatAgent to an OpenAI model, attaches a search tool, and sends multi-step prompts. Its main building blocks include autonomous agents, agent societies for role-playing and workforce-style collaboration, model backends, tool integrations, stateful memory, persistent storage, data loaders, retrievers, code and command interpreters, runtime management, benchmarks, and human-in-the-loop oversight. The documented design principles are evolvability through data and environment interaction, scalability for large agent systems, stateful memory for multi-step tasks, and code-as-prompt readability. The README describes support for dynamic communication and simulations containing up to a million agents. CAMEL also provides synthetic-data workflows for chain-of-thought generation, self-instruct, Source2Synth, and self-improving CoT data, along with guides for fine-tuning and publishing datasets. It links Hugging Face datasets covering AI society conversations, code, math, physics, chemistry, and biology. RAG and Graph RAG cookbooks help agents retrieve external knowledge. Documented applications include collaborative role playing, multi-agent workforces, report and knowledge-graph generation, hackathon judging, customer-service Discord bots, website ingestion, video analysis, and world simulation through the OASIS project. The repository includes examples, documentation, a Colab demo, research links, and community support channels.