OpenEnv: Agentic Execution Environments

An end-to-end framework for creating, deploying and using isolated execution environments for agentic RL, with a simple Gymnasium-style API.

One API for every environment

reset(), step() and state(), sync or async, over a WebSocket.

Isolated and deployable

Each environment is a Docker image that runs locally, on a cloud sandbox, or as a Hugging Face Space.

40+ environments

Games, coding sandboxes, browsers, finance, simulators and more, in the catalog.

Train with your framework

TRL, Unsloth, SkyRL, ART, Oumi, torchforge, Miles and more. See Training with OpenEnv.

Train real coding agents

Harbor runs Claude Code, Codex, OpenCode and other harnesses, and captures their tokens for RL.

Serve tools to agents

MCP environments expose their tools over /mcp in production mode.

Where to start

Getting Started: install OpenEnv, connect to an environment and run your first step.
Train an agent: pick a way to train and a framework. Harbor captures coding agents that run their own loop, for training.
Build your own environment, then deploy it to Hugging Face Spaces.
Explore environments: browse the catalog.

The tutorials include a 5-part Getting Started series that needs no GPU, and the Concepts pages explain how the pieces fit.

Contributing

OpenEnv is openly governed by a technical committee that coordinates project direction, RFCs and releases through the public GitHub repository. The charter explains how it works, and the README lists the committee members and supporters. Bug reports, feature requests and new environments are welcome as issues or pull requests, see Contributing. For the changelog, see GitHub Releases.

OpenEnv is in early development, so APIs may still change. Bug fixes are welcome. For larger changes, open or claim an issue first so the change can be discussed.

Update on GitHub