Building the World Layer Where Machines Learn to Think.

Our approach starts with games because they they’re the cleanest compression of reality available to humans; they distill the messy complexity (the uncertainty, the physics, the social dynamics) of the real world into tight loops of perception, reasoning, and action. They let us train for capabilities necessary for the next generation of intelligent systems: long-horizon planning, real-time control, coordination, robustness, and generalization.

Our mission is to become the world layer for machine intelligence that helps close the gap between benchmark performance and real-world capabilities through a continuously evolving supply of environments, trajectories, and human reasoning for LLM/VLM agents and robots.