Welcome to OpenRA-RL

OpenRA-RL is now live! We're building a Gymnasium-style reinforcement learning environment for the OpenRA real-time strategy engine.

What is OpenRA-RL?

OpenRA-RL lets you train AI agents that play full games of Red Alert through the OpenRA engine. The environment provides:

• Rich observations: 9-channel spatial tensor, per-unit stats, economy and military data
• 21 action types: Move, attack, build, train, guard, transport, and more
• Real-time bridge: gRPC connection running at ~25 ticks/second
• Multiple agent types: Scripted bots, LLM agents, and RL policies

Getting Started

Check out the Getting Started guide to set up your first agent, or explore the Architecture docs to understand how everything connects.

OpenRA-Bench

We're building OpenRA-Bench — a community leaderboard for comparing agent performance with verified replay data. Stay tuned for the launch.