drewgent-atropos-environments

What problem does it solve?

Drewgent-Atropos Environments provide a structured approach to building, testing, and debugging reinforcement learning environments for the Atropos training framework, enabling reproducible experiments and early validation of environment logic.

Core Features & Use Cases

• Scaffolds the DrewgentAgentBaseEnv interface implementation, including agent loop orchestration, tool resolution, and reward verification hooks.
• Documents file organization, evaluation workflows, and the three CLI modes (serve, process, evaluate) to accelerate development and benchmarking of RL environments.
• Guides integration with monitoring and logging (e.g., wandb) and supports end-to-end agentic evaluation with tools in a controlled sandbox.

Quick Start

Configure and implement a new Drewgent RL environment following the DrewgentAgentBaseEnv API, then run it in process or evaluate mode after selecting an inference backend.