American startup teaching robots to speak while outsourcing foreign mechanics
Headquarters
San Francisco
Employee
—
Models here
1
Listings
0
Lightberry was founded in San Francisco in 2025 and comes from conversational AI, not mechanical engineering. The company positions itself as a behavioral layer for humanoid robots—the part that listens, speaks, and derives actions from what it hears. In August 2026, it introduced the Lumi, its first proprietary device.
At its core is Lightberry OS, an operating system that deliberately divides the work: perception and motion control run locally on an NVIDIA Jetson Thor, while speech understanding happens in the cloud. With the Lumi, Lightberry combines this software with the mechanics of a Unitree G1, supplemented by its own three-axis neck, stereo cameras, and an eight-channel microphone array. Assembly takes place in San Francisco.
Lightberry exemplifies a division of labor taking shape across the industry: mechanics increasingly come as off-the-shelf components from Asia, while value is created in perception, speech, and behavior. The Lumi, priced at nearly 40,000 USD plus an annual software fee, targets companies and institutions. It remains unclear how American import restrictions on foreign robots will affect this model.
Lumi
Humanoid · Interaction and Reception
Lightberry OS
Software · Speech and Behavior
Founded in San Francisco, accepted into Y Combinator batch F25.
Introduction of first proprietary robot Lumi on August 26, reservations opened.
No distribution partner known in Switzerland. Reservations have so far been handled directly through the manufacturer.