Runtime (YC F26) is a robot policy for teams deploying robots that still need a new fine-tune or a teleoperation session for each task. It turns what the robot sees into a simulation, searches options there, acts on the one that works in the real world, and keeps the simulation in sync with the scene. The launch is aimed at companies putting more robots into the field, at a point when the founders say simulation is fast and accurate enough to guide action, not only training and evaluation. The launch video shows a robot threading a rope with no fine-tuning and no teleoperation. Runtime says its experiments match or beat leading robot foundation models on real tasks including rope threading and egg grasping. The company is now speeding that loop so it can run live, and early access opens in January. Liam Zhong, Yim Ying, and Chen Peng founded the San Francisco company out of the University of Pennsylvania. They describe the group as coming from Penn's GRASP Lab, with experience in computer vision, control, reinforcement learning, and robot foundation models, and with papers and reviews at CVPR, ECCV, and NeurIPS. Zhong is co-founder and CTO.