What this is

Faïence is a free, open-source implementation of the rules of Azul, the tile-laying game by Michael Kiesling. It is a fan project for research, with its own code and artwork, not affiliated with or endorsed by the game's publishers.

All of it is open source (MIT): code and training logs on GitHub. If you like this work, a star on the repo (or a hello in the issues) is always appreciated.

Faïence began as a machine-learning experiment: Ludometer, a research framework that measures how good a board game is from the shape of an AI's learning curve. Somewhere along the way the experiment turned into a pleasant opponent, so we are sharing it, playable and free. Nobody earns a cent from this page: no ads, no account, nothing to buy.

The opponent is a neural network that learned the game from scratch by playing against itself. Your browser downloads it once and runs it locally.

This is a research project, and the games are the research material: when you play, you help the net train. So when a game ends (or is abandoned), the page sends an anonymous record of it to be used as training data: the moves, the tiles that were dealt, which net you played, and the score. No name, no account, no location. This is experimental, and there is a switch in Settings if you would rather not. Every record is public, so you can read everything we collect.

Beyond that, the only thing that leaves your browser is an anonymous, cookie-free tally of visits, games dealt, and how finished games went. It is public too.