CAMEL

Built around the role-playing technique from its NeurIPS 2023 paper, CAMEL is a research-grade multi-agent framework whose stated mission is finding the scaling laws of agents, supporting simulations with as many as 1 million communicating LLM agents. Agents get stateful memory for multi-step interactions, specialized toolkits such as search integrations, and support for a wide range of model backends, and the framework doubles as a synthetic data generation engine for producing training and evaluation datasets. The camel-ai organization maintains a notable ecosystem on top of the core library, including the popular OWL agent project, the OASIS social simulation platform, CRAB, and Loong, so it effectively serves as shared infrastructure for agent research. Installation is pip install camel-ai, cookbooks cover data generation, Discord bots, and infrastructure automation, and documentation lives at docs.camel-ai.org. All code is Apache-2.0 with around 17.5k GitHub stars and more than 2,200 commits, and typical deployments pair it with hosted LLM APIs, though open model backends are supported as well.