environments

What problem does it solve?

RL development often requires custom, reusable environments with consistent interfaces to train and evaluate agents, but building and maintaining these environments across experiments is error-prone and time-consuming.

Core Features & Use Cases

• Env protocol, EnvGroupBuilder, and RLDataset to structure single- and multi-turn interactions.
• Prebuilt references and examples like Math RL env, Harbor RL env, and message-based env templates to accelerate integration.
• Use cases span academic research, prototype experimentation, and production-style RL pipelines with reproducible lifecycles.

Quick Start

Create a MyEnv by implementing the Env interface, wire it into an EnvGroupBuilder, and plug it into your training loop.