DouZero uses self-play deep reinforcement learning to train AI agents capable of playing DouDizhu, a shedding-type card game characterized by competition, collaboration, imperfect information, large state space, and highly variable legal actions. Developed by Kwai Inc.’s AI Platform, the framework enables agents to learn competitive strategies purely through gameplay against themselves, achieving strong performance without human data. The project provides pretrained models and an online demo for playing against or observing the trained agents.