KataGo is an open-source Go engine developed through self-play training and community research. It offers strong play and can predict scores and territory, handle handicap games, and support different board sizes and rules using the same neural network. The project focuses on improving playing strength and training efficiency through evolving engine algorithms, network architectures, and data-generation techniques.

This site hosts a public distributed training run, where volunteers contribute computing resources to generate self-play data and continue training KataGo beyond its earlier official run. It also provides trained networks, including models for general play, rating games, and large-board use, alongside archives, releases, and information about the training effort. Contributors can create an account and use spare GPU capacity to support the run; compatible KataGo versions are required as the project updates its training formats and models.

People who want to use the engine can download KataGo releases and networks from the site.