Physical Intelligence is opening the Pi API beta to more robotics teams, giving them a hosted way to adapt its general-purpose robot foundation models to their own hardware and jobs. A developer uploads demonstration data, launches fine-tuning on compute the company manages, and calls the tuned model through a hosted gateway to run inference on the robot. Access is waitlisted, and applicants are asked to say what they want to build. The company says the model currently exposed through the API already fine-tunes with less partner data, and that it can update the underlying models without teams rebuilding their own training stack. The beta follows work with partners across industries, which the company describes as ranging from strawberry harvesting to server assembly. In the launch materials, Actor's autonomous wheel loader scoops material into a dump truck in a commercial deployment with no operator inside the machine. Almond, which builds the Axol dual-arm robot, obtained a model after four days of API iteration that deburrs spacers used in Axol's own assembly, according to Physical Intelligence. Other named users include AgRobotics, Chef Robotics, Formic Robotics, CloudChef, and teams working on food preparation, heavy equipment, and manufacturing. Most of the shown applications ran on cloud-hosted inference, including sites in Japan, Switzerland, and the United States, though the company noted connectivity problems on some deployments. Physical Intelligence, based in San Francisco and founded in 2024, develops learning models for controlling robots across different hardware. Co-founders include Karol Hausman, Sergey Levine, Chelsea Finn, Lachy Groom, Brian Ichter, Quan Vuong, and Adnan Esmail. The company already releases open versions of models such as π0 and π0.5. The API is the first broader release of the partner workflow, aimed at teams that want to iterate on their own robots and data while Physical Intelligence runs the training and serving infrastructure.