Skip to content

RLMatrix Documentation

RLMatrix provides a comprehensive reinforcement learning framework for C# developers with performance exceeding Python alternatives.

Key Features

  • Comprehensive Algorithm Library: Includes PPO, DQN with all popular modifications (up to C51 and DQN Rainbow), GAIL, and more algorithms on the way

  • Revolutionary DRL Workflow: Thanks to C# source generation in the toolkit, you can focus on your domain problem rather than wrestling with complex API requirements - simply add attributes to your code and the toolkit generates all the reinforcement learning plumbing automatically

  • Multi-Head Support: Handle continuous, discrete, and mixed action spaces simultaneously in a single agent

  • RNN Integration: Enable recurrent neural networks with a simple option toggle for handling sequential or partial observability problems

  • Pure C# Implementation: Built entirely in C# with TorchSharp backend, providing native performance and complete type safety

  • Game Engine Ready: Battle-tested in Unity and Godot.

  • Superior Performance: Faster and more stable than Python’s stable-baselines, ml-agents, and Godot RL agents

  • Multi-Environment Training: Scale learning across parallel (optionally networked) environments

  • Industrial-Grade Distributed Training: High-performance, fault-tolerant networked architecture ready for large-scale reinforcement learning deployments

  • Transparent Source Code: Clean, well-documented codebase that’s easy to extend or customize. Reinforcement Learning with Dependency Injection!

  • Production Reliability: Designed for stability in long-running training sessions with failure tolerance

    Get Started with RLMatrix →