Crab

This framework is developed for creating environments for agents based on large language models. It supports the creation of environments for cross-platform testing, which allows deploying them on various systems, including in-memory environments, containers, virtual machines, or distributed physical machines.

The target audience includes researchers, developers, and specialists who require building and evaluating environments for LLM agents.

The framework provides a convenient interface based on Python for defining environments and agent actions, ensuring flexibility for various scenarios.

Key features include support for cross-platform and multi-environment deployment, a unified interface for accessing environments, and a native configuration layer in Python.

The framework includes an innovative benchmarking suite that provides detailed evaluation metrics.

The framework is used for benchmarking LLM agents, testing cross-environment interaction, processing multimodal data, and simulating environments for agents.