Ali Attar on Lightberry's natural-language OS for humanoid robots and a just-closed deal with Unitree

Summary

Ali Attar, co-founder of Lightberry, is building what he describes as an operating system layer for humanoid robots — software that lets anyone interact with a robot through natural language rather than code. The pitch is straightforward: you can buy a humanoid robot from any of roughly 50 manufacturers today, but out of the box it can't speak, can't be taught anything, and can only be controlled by writing code. Lightberry sits between the hardware and the user, handling conversation, personality, and autonomous behavior.

The company's most concrete commercial development is a deal closed with Unitree, signed the week before this recording. Attar describes Unitree as holding roughly 90% of global humanoid robot market share and, critically, as the only manufacturer actually shipping at scale. American players like Figure and Optimus are not yet in market, and Attar says Figure's partnership with OpenAI may have fallen through — though he flags that as unconfirmed rumor. The Unitree deal positions Lightberry to sell its software bundled with Unitree robots across the US, with European and American hardware partners also in early discussions.

The go-to-market logic runs through manufacturers, not end users. Hardware companies are racing on motors, actuators, and form factors and have little bandwidth for the interaction layer. Lightberry fills that gap and sells the software as part of the robot purchase. Attar's first customer was a toy company. His building's landlord pre-ordered four robots at $60,000–$70,000 each for security after a single hallway conversation — the use case being deterrence and perimeter monitoring rather than physical intervention.

On form factor, Attar expects a Cambrian explosion rather than a single winner. Wheeled robots for delivery, small bipedal droids for homes, and humanoids for people-facing roles — shop assistants, event staff, reception. His argument for humanoids in public spaces is simple: they look like people, so they slot naturally into roles built around human interaction. A Lightberry-powered Unitree robot MCed the YC Demo Day event for the full day, and three weeks earlier a humanoid ran autonomously alongside event staff for 10.5 hours at the a16z summit.

Lightberry runs a hybrid compute pipeline — primarily cloud for model quality, with an offline fallback so the robot can still hold a conversation if connectivity drops. Attar is not trying to make robots indistinguishable from humans. The design target is closer to C-3PO: clearly a robot, but socially fluent enough to follow norms, make eye contact, and behave appropriately in context.

The team is three founders. Attar says the seed round closed early, and the Unitree deal puts them close to a Series A milestone.