bcollazo/catanatron

Settlers of Catan Bot Simulator and Strong AI Player

What it solves

Catanatron provides a high-performance simulator and AI player for the board game Settlers of Catan, enabling the development and testing of strong AI agents. It addresses the need for a fast simulation environment where thousands of games can be played in seconds to iterate on AI strategies.

How it works

The project implements the core game logic of Settlers of Catan as a Python library. It offers three primary ways to interact with the game:

  • Command Line Interface (CLI): A tool called catanatron-play allows users to run large-scale simulations and export game data in formats like JSON for analysis.
  • Graphical User Interface (GUI): A Docker-based web UI allows users to watch, inspect, and play against the AI.
  • Gymnasium Interface: An OpenAI/Gymnasium environment wrapper allows reinforcement learning agents to interact with the game, providing observations, rewards, and valid actions.

Who it’s for

  • AI researchers and developers interested in game AI and reinforcement learning.
  • Developers wanting to build and test Catan bots.
  • Users who want to play against a strong Catan AI.

Highlights

  • High Performance: Capable of simulating thousands of games per second.
  • RL Ready: Integrated with the Gymnasium interface for easy reinforcement learning training.
  • Data Generation: Built-in CLI tools to generate datasets from simulated games.
  • Web UI: Includes a Docker-based interface for visualizing game play.

Related

  • Project
    Eclectic-Sheep/sheeprl

    A scalable PyTorch reinforcement learning framework accelerated by Lightning Fabric that decouples RL algorithms from their environments for greater flexibility.

  • Project
    neiii/bridle

    A unified configuration manager for AI coding assistants that allows users to manage profiles and install skills or MCPs across multiple tools from a single interface.

  • Project
    bytedance/InfiniStore

    InfiniStore is an open‑source, high‑performance key‑value store that lets LLM inference clusters share and reuse KV attention caches across nodes, reducing GPU memory pressure and latency. It works with vLLM today, supports TCP and RDMA networks, and can also be used as a generic KV store for other LLM services.

  • Project
    discopy/discopy

    A Python toolkit for computing with string diagrams, enabling the representation and evaluation of monoidal categories as tensor networks or functions.