Mbodi AI

Mbodi AI is a New York-based startup building embodied-AI software that enables industrial robots to learn and operate like human workers. The company was founded by Xavier (Tianhao) Chi, a former tech lead of Google Public DNS, and Sebastian Peralta, a physicist, roboticist, and deep learning researcher who was a core engineer at Google Public DNS and conducted graduate research at UPenn's GRASP Robotics Lab.

Mbodi's platform lets users teach robots new manipulation skills directly on the factory floor through natural conversation and physical demonstration, with no programming, specialists, or production downtime required. It combines generative AI with agent orchestration to translate everyday language and quick demos into precise, reliable robot actions. The system adapts in real time, refining its behavior from every interaction and adjusting instantly to new inputs, SKUs, or environments, and is engineered for production rather than one-off research demos, with the company citing 99 percent uptime and predictable performance in real-world conditions. Knowledge learned by one robot transfers across fleets and facilities, so capability scales as shared learning.

Mbodi joined Y Combinator's X25 batch and was named, alongside T-Robotics, a winner of ABB Robotics' AI Startup Challenge, with ABB describing its technology as a major advancement in making robotic automation accessible. The company was recognized among The Robot Report's RBR50 list of innovative robotics companies and highlighted as one of the most exciting AI agent startups at Y Combinator's Demo Day. Mbodi targets industrial automation use cases where traditional reprogramming, consultants, and downtime cannot keep pace with evolving product lines and customer needs.