Liquid AI builds personal AI around device-level context

Liquid AI is pushing a simple idea: if AI is meant to feel personal, it should live closer to the person. Jeffrey Li, the company’s chief operating officer, said the real home for that kind of system is the edge, where devices can see more of a user’s day-to-day context than a remote cloud service ever will.

That sounds neat until you hit the hardware wall. On-device AI has to fit inside fixed compute, which means the usual cloud assumption — that resources can expand whenever needed — stops working. Li said that changes how models, agents and their software harnesses are built, because context can’t just grow forever in a giant text file on a small device.

Liquid AI’s answer is Liquid Context, a layer optimized for Snapdragon processors that sits between the model, the agent and the hardware. It uses signals from devices to figure out who the user is and what they’re trying to do, and then decides what information to keep and how to compress it. The devices in your pockets, Li argued, are the best place to collect that signal: phones, wearables, watches, PCs and cars.

The company is already taking that thinking into the car with Mercedes-Benz Group AG. And Li said the next step is making these systems improve after deployment, not just ship and freeze. That means observability loops and continuous improvement loops that can help both the model and the harness self-heal, adapt and personalize through normal use.