Agent Gym: A Framework for Continuous Evaluation and Evolution of LLM Agents Through Human-in-the-Loop Feedback
Agent Gym is a modular framework designed to fix LLM agents that need to change after deployment. It wraps existing agents in a loop allowing continuous evaluation, correction, and learning guided by human experts. The system uses configuration files for knowledge and a structured pipeline for acting and validation, enabling safe, iterative improvement without rewriting the agent's core code.