Crab is a comprehensive framework developed by Camel AI for building and benchmarking environments specifically tailored for large language model (LLM) agents. The platform supports the creation of cross-platform environments, enabling deployment across in-memory systems, Docker-hosted environments, virtual machines, or distributed physical machines. Crab offers an easy-to-use Python-centric interface for defining agent environments and actions, making it highly flexible for various use cases. Additionally, it includes a novel benchmarking suite that provides fine-grained evaluation metrics to assess the performance of LLM agents accurately.