OpenEnv
BSD-3-Clause framework for building and serving isolated agentic RL environments over a Gymnasium-style reset/step/state API.
Tags
Description
Established open project (canonical repo openenv/OpenEnv; not archived; created 2025-10-01; not presented as a 2026 launch). SPDX BSD-3-Clause; LICENSE copyright 2026, Hugging Face, Inc. PyPI openenv 0.13.0 verified 2026-11-13 (not yanked; wheel uploaded 2026-11-12T06:33:48Z, which is 2026-11-12 07:33 Europe/Stockholm). GitHub latest release v0.13.0 was published 2026-11-12T06:34:01Z (2026-11-12 07:34 Europe/Stockholm). README install is pip install openenv. Docs at https://huggingface.co/docs/openenv/index and https://huggingface.co/docs/openenv/main/en/index (both HTTP 200). README and the docs index describe a Gymnasium-style reset(), step() and state() API over WebSocket/HTTP, Docker-packaged environments that can run locally, on a cloud sandbox, or as a Hugging Face Space, and “40+ ready-to-use environments” in the environment catalog (https://huggingface.co/docs/openenv/environments). An exact catalog census was not taken beyond that documented claim. Integrations named in the README include TRL, Unsloth, SkyRL, ART and others. Environments may execute untrusted agent or task code in containers — treat them as potentially destructive sandboxes. GPU, model-weight and hosting costs are separate.
Agent access
Related
- TRL: Hugging Face Transformer Reinforcement Learning library for post-training methods such as SFT, DPO, PPO and GRPO on top of transformers.
- Unsloth: Open-source LLM fine-tuning library and desktop app for running and training models locally, with a dual Apache-2.0 / AGPL-3.0 license split.
- SkyRL: Modular reinforcement-learning stack with a training engine, long-horizon agent layer and Gymnasium-style tool-use environments for LLMs.
- ART (Agent Reinforcement Trainer): Open-source Python framework for training multi-step agents with GRPO, with an OpenAI-compatible client and a separate GPU training server.
- verl: Distributed reinforcement-learning training library for LLMs, with GRPO/PPO dataflows and pluggable training and rollout engines for agentic post-training.
- Harbor: Framework from the Terminal-Bench creators for evaluating and optimizing agents and models in sandboxed environments, run in parallel across cloud sandbox providers.
- verifiers: Prime Intellect’s MIT library for creating LLM reinforcement-learning environments and evals, with a matching PyPI package.
- NeMo Gym: NVIDIA Apache-2.0 library for building, evaluating and scaling agent environments, verifiers and harnesses, including sandboxed code-execution tasks.
- prime-rl: Prime Intellect’s Apache-2.0 asynchronous RL training framework, installed from git/uv (no verified PyPI package) and integrated with verifiers environments.
Guides & comparisons
- RL environment libraries compared — NeMo Gym, OpenEnv and Verifiers (Comparison): Compare NeMo Gym, OpenEnv and Verifiers as libraries for writing and running RL environments: authoring model, isolation, documented trainers, install paths, licences and current versions. Distinct from Harbor/AgentEnv/Seahaven and from the agent RL stacks comparison.
Sources
- https://github.com/openenv/OpenEnv
- https://github.com/openenv/OpenEnv/blob/main/LICENSE
- https://github.com/openenv/OpenEnv/blob/main/README.md
- https://pypi.org/project/openenv/
- https://huggingface.co/docs/openenv/index
- https://huggingface.co/docs/openenv/environments
- https://github.com/openenv/OpenEnv/releases/tag/v0.13.0
Machine-readable
- JSON
- Markdown
- Source file on GitHub (edit via pull request)