Liquid AI Brings Shared AI-Agent Context to Snapdragon Chips With Qualcomm

An AI assistant can be more useful when it remembers your routines. Liquid AI wants that memory to live on a Snapdragon device and serve the agents you choose, rather than belong to just one of them. At Snapdragon Summit on September 23, Liquid AI announced a collaboration with Qualcomm that optimizes its Liquid Context software for the Hexagon chip component that runs AI workloads.

From device signals to a shared profile

With permission, Liquid Context is designed to learn from signals on a device and keep an up-to-date picture of a person’s routines, preferences and needs. That personal context is built and maintained locally. A connected agent receives relevant parts according to the user’s permissions. Liquid says the agent can be its own Liquid Agent or one from another company.

The Hexagon NPU, Snapdragon’s processor for AI tasks, is meant to handle the recurring work in the background. Liquid says this design does not require a cloud model to process every update. That is different from saying every agent stays on the device: the company says agents using the permitted context may run locally, in the cloud or across both. The distinction matters when judging the promise of an on-device profile.

The handoff Liquid wants to enable

Liquid’s examples show what it wants agents to do with that shared understanding, not what it has confirmed in a shipping device. In one illustrative scenario, a message says a child is sick and needs an early school pickup. Liquid Context supplies relevant family priorities; an agent checks the calendar, identifies meetings that can move and drafts emails for the user to approve. The context provides background, while the agent handles the task.

Device makers get a choice, not a rollout date

For manufacturers, Liquid is pitching Context as a built-in capability that could work with their choice of agents and services. It is also offering a separate option: manufacturers can evaluate Liquid Agent, which runs on Liquid’s LFM2.5-2.6B model and has been optimized for the Hexagon NPU. Liquid says that agent can be tailored to a manufacturer’s hardware, services, interface and brand. Using the context layer does not, in the company’s proposed setup, require choosing Liquid’s agent.

The next step belongs partly to those manufacturers. Liquid says it and Qualcomm are exploring further work to make the context layer easier to offer as a built-in feature, while The Next Web notes that neither company named a device maker. The announcement also leaves users without a specific product or date to look for. For now, the concrete change is an optimization for Snapdragon hardware and a proposed way for several agents to draw on one permissioned profile—not a confirmed experience on a named phone, watch or car.