Trinity-RFT decouples the RFT process into three coordinated components: Explorer for agent-environment interaction data generation, Trainer for model weight updates via loss minimization, and Buffer for data pipeline management throughout the RFT lifecycle. It provides modular, plug-and-play components allowing non-invasive customization for both agent application developers and reinforcement learning researchers.