CAMEL

CAMEL is one of the first and most influential open-source multi-agent frameworks, built by a research community of more than 100 contributors to study how large societies of LLM agents behave and scale. Agents with different roles communicate, use tools and remember state to solve tasks together, from data generation to full world simulations. The project has earned about 17,900 stars on GitHub and underpins well-known research systems such as OWL and OASIS.

Overview

The framework is organized around four design principles: evolvability, so agent systems improve through generated data and environment interaction; scalability, targeting systems of up to a million agents; statefulness, with persistent memory across long interactions; and code-as-prompt, keeping code readable to agents as well as humans. Its modular stack covers agents and agent societies, dozens of model backends, toolkits for search, code execution, browsers and more, memory and storage layers, retrievers for RAG, benchmarks and interpreters. Extensive cookbooks and real-world use cases cover synthetic data generation, task automation, infrastructure workflows and multi-agent simulations of social platforms.

Key Features

• Multi-agent framework designed for scale, with role-playing agent societies
• Modular toolkit: agents, models, tools, memory, storage, retrievers and benchmarks
• Synthetic data generation pipelines including Self-Instruct and chain-of-thought data
• Supports many model backends, from OpenAI and Anthropic to open and local models
• Foundation of notable research systems including OWL and the OASIS social simulation
• Rich cookbooks for automation, RAG, document intelligence and research workflows

💳 Pricing

CAMEL is completely free and open source under the Apache-2.0 license. You run it yourself and pay only for the model APIs your agents use.

Who Is It For

CAMEL is for researchers and developers exploring multi-agent systems, synthetic data generation and agent societies, especially where experimentation and extensibility matter more than a managed platform.