hermes-atropos-environments

To build an RL environment, extend the Atropos BaseEnv with HermesAgentBaseEnv features to enable tool resolution, deterministic agent loops, and reward computation for full RL evaluation. This framework provides the necessary environment scaffolding and lifecycle methods.

Running sandboxed tool-based interactions requires integrating HermesAgentLoop with ToolContext within your Atropos environment. This setup enables deterministic agent loops and robust tool execution for RL research and deployment.

Yes, you can use wandb logging for Atropos RL environment evaluation by implementing the wandb_log lifecycle method. This allows you to track metrics during full evaluation runs and benchmark model performance.

Yes, the framework supports serve, process, and evaluate CLI modes. These workflows allow you to generate data, validate agent behavior, and benchmark models using full HermesAgentLoop playthroughs with tool calls.

A configurable Hermes Atropos environment requires setup, get_next_item, format_prompt, compute_reward, evaluate, and wandb_log lifecycle methods. These methods ensure proper environment initialization, data processing, and evaluation.