Best way to use rllib with player vs player games

I wonder what is the best way to use rllib to train agent in player vs player turns games like chess.
What is the easiest way to implement training from scratch by playing agent against itself.
I will be gratefull for any suggestions.

I asked a similar question a couple weeks ago: Best training algo for turn based board game?

Check out also what I’ve done so far at

Check out also the RoundRobinMultiAgent example environment: ray/ at 929946271db1f7cd84716a2c1850e8bec90841c3 · ray-project/ray · GitHub

@aronsar Thank Yoy for the answer