huggingface/OpenEnv

Health v2 · maintenance only

Adoption not part of the score

2521 stars · 428 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

OpenEnv is a Hugging Face framework for creating, deploying, and consuming isolated execution environments used in agentic reinforcement learning post-training of LLMs. It provides Gymnasium-style step/reset/state APIs, container-based HTTP environment servers, auto-discovery clients, a CLI for scaffolding and deployment, and composable rubrics for reward design.

Use cases

• create custom environments for RL post-training of LLMs
• train agents with GRPO in gym-style environments
• deploy sandboxed execution environments for agentic RL
• integrate TRL or torchforge trainers with external environments
• package an RL environment as a Docker container and serve it over HTTP
• find prebuilt environments for coding, games, or web-browsing tasks
• design reward functions and rubrics for RL training

When to choose

• You are doing RL post-training (e.g., GRPO) of LLMs and need a standardized environment interface
• You want Gymnasium-style step/reset APIs that plug into TRL, torchforge, or SkyRL
• You need secure, containerized, HTTP-served environments for distributed or remote training
• You want to publish environments to Hugging Face Spaces and auto-discover them via AutoEnv/AutoAction

When to avoid

• You run classic single-machine RL research (Atari, MuJoCo) where plain Gymnasium already suffices
• You need an inference-time agent framework rather than training environments
• Your stack is not Python-centric
• You require a mature 1.0 release; the project is still at v0.4.x with some docs in progress

Facets

framework · maturity active

reinforcement-learning agent-framework http-server mcp cli reinforcement-learning large-language-models machine-learning developer-tools python cloud self-hosted gymnasium rl-environments post-training grpo reward-design rubrics sandboxed-execution environment-server auto-discovery fastapi hugging-face trl-integration containerized-environments ai-agents docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="huggingface/OpenEnv")