AgentGym
Agent training and evaluation suite exposing diverse interactive environments through a common HTTP interface and Python clients.
Tags
Description
Public MIT-licensed code with the agentenv client (pip install agentenv). Environment servers expose createEnv, observation, available_actions, step and reset; install each selected agentenv-* environment separately. Covers web, household, games and tool/SQL environments; AgentController connects agents for evaluation, data collection and training. Links AgentEval and AgentTraj-L datasets, whose reuse terms must be checked separately; this entry licenses the code, not every bundled environment or external dataset. Model and environment compute are not included.
Agent access
Related
- WebArena: A realistic web environment for building autonomous agents, with benchmark tasks used to evaluate web agents.
Sources
- https://github.com/WooooDyy/AgentGym
- https://github.com/WooooDyy/AgentGym/blob/main/LICENSE
- https://pypi.org/project/agentenv/
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)