SKU RLE-1305 · Sold by External
AgentGym (14 environments)
Product specifications
| SKU | RLE-1305 |
|---|---|
| Data type | RL environments |
| Volume | 14 environments |
| Size on disk | Not published by source |
| Format | Python + HTTP env servers (agentenv) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | MIT License |
AgentGym is a framework for training and evaluating LLM-based agents across 14 diverse environments spanning web navigation (WebShop, WebArena), text games (MAZE, Wordle), house-holding (ALFWorld), embodied science (ScienceWorld/SciWorld, BabyAI), digital games (TextCraft), tool use (Weather, Movie, Academia, Sheet, TODOList), and programming (BIRD SQL). All environments are served behind a unified HTTP/ReAct interface (createEnv, observation, available_actions, step, reset) with concurrency support. It ships expert trajectory sets (AgentTraj, AgentTraj-L) and the AgentEval benchmark on Hugging Face for cross-environment evaluation.