OpenEnvironment is an RL-for-games environment construction toolkit: generate and mutate game-ready world state and assets via an API so you can run training loops against consistent, replayable environments.
Today, the concrete environment module in this repo is terrain/world layout generation (heightmaps, splatmaps, optional voxel meshes). The longer-term goal is a broader environment builder (scenarios, constraints, evaluation signals) suitable for RL workflows.
Everything lives under OpenEnvironment/. The repo root only contains launcher scripts and this README.
# 1. Backend
cd OpenEnvironment/server && uv sync && cd ../..
echo CEREBRAS_API_KEY=your_key > OpenEnvironment/server/.env
.\start-backend-uv.ps1 # http://localhost:8001
# 2. Frontend (new terminal)
cd OpenEnvironment/web && pnpm install && cd ../..
.\start-frontend.ps1 # http://localhost:5173Needs Python 3.10-3.12, Node 18+, uv, pnpm, and at least one LLM API key (Cerebras, Together, or Gemini).
Without an API key the system falls back to regex parsing (no LLM parsing features).
See OpenEnvironment/SETUP.md for details, OpenEnvironment/ENV_TEMPLATE.md for all env vars.
- Stateful environment iteration: generate/modify/reset/regenerate from a persisted JSON state
- Generation outputs: heightmap + splatmap (+ optional voxel mesh export)
- Reference resolution: scene graph entities (“the dunes”, “first mountain”)
- Evaluation signals: composition/texture metrics + rubric scoring
- Two clients: REST API + a React/Three.js viewer
Command or template
-> parse to actions (narrative -> semantic -> regex)
-> apply actions to state (add/modify/remove)
-> build terrain from state (heightmap + splatmap)
-> evaluate (metrics/rubric)
-> return state + asset paths
API (backend on http://localhost:8001)
| Endpoint | Method | Purpose |
|---|---|---|
/api/generate |
POST | Generate terrain from text |
/api/modify |
POST | Modify existing terrain |
/api/state |
GET | Current terrain state |
/api/reset |
POST | Reset terrain |
/api/templates |
GET | List templates |
/api/templates/{id}/apply |
POST | Apply a template |
/api/mcp/tools |
GET | List MCP tools |
/api/status |
GET | Server/API key status |
Full reference: OpenEnvironment/docs/usage/api.md
OpenEnvironment/
server/ Python backend (FastAPI)
web/ React + Three.js frontend
docs/ Documentation (vision, architecture, usage, development)
SETUP.md Setup instructions
ENV_TEMPLATE.md Env var template (copy to OpenEnvironment/server/.env)
FastAPI, NumPy, SciPy, Pillow (backend). React 18, Vite, Three.js, Zustand (frontend). Cerebras/Together/Gemini for LLM.
- Start here: OpenEnvironment/README.md
- Docs index: OpenEnvironment/docs/README.md
- Vision: OpenEnvironment/docs/vision/RL_FOR_GAMES.md
MIT
