The Agent Reinforcement Trainer (ART) library has been developed to simplify training large language models (LLMs) through a user-friendly interface. With no callbacks required, it offers an OpenAI API-compatible endpoint, allowing it to replace any proprietary API counterparts. Users can customize their training processes by setting their own reward structures after collecting responses from the inference API, enabling continuous improvement until the desired performance is reached. The developers highlight a significant success with training an email research agent, showcasing the framework's potential in practical applications of reinforcement learning (RL). This flexibility in model training is expected to encourage users to create state-of-the-art solutions tailored to their specific needs. The developers are open to engaging with the community for any questions about the framework, emphasizing collaboration and knowledge sharing.