AI assistants can answer questions, summarize documents, and increasingly take actions on our behalf. But there is still something holding them back: they usually don’t know enough about us. Liquid AI thinks the solution is to give AI a persistent memory of your life, and it is working with Qualcomm Technologies to make that possible directly on devices powered by Snapdragon.
Announced at Snapdragon Summit 2026, Liquid Context is an on-device context layer that has been optimized for Snapdragon processors and the Qualcomm Hexagon NPU. It is designed to continuously build an understanding of a user from permitted signals generated by their devices. That context can then be made available to AI agents running locally, in the cloud, or through a combination of both.
SEE ALSO: Microsoft just made its smallest Surface PCs dramatically more powerful with Snapdragon X2 Plus
This is potentially much more interesting than simply sticking another chatbot on a laptop or smartphone. An AI that remembers your routines, preferences, priorities, and previous activity has a much better chance of actually being useful. It is also a little creepy.
Liquid AI says personal context is built and maintained locally, with information made available to connected agents according to permissions set by the user. Running the system on the Hexagon NPU means device signals can be processed continuously without sending every update to a cloud-based model. That fits neatly with Qualcomm’s broader push toward personal AI running across the devices people use every day.
The examples Liquid AI provides show just how personal this could become. If a parent receives a message saying their child is sick and needs to leave school early, an AI agent could understand the family situation, inspect the parent’s calendar, determine which meetings can be moved, preserve an important commitment, and prepare emails to reschedule everything else.
Another example follows someone through a conference. Liquid Context could remember what happened during the day, understand that the user likes posting recaps to LinkedIn, access approved photographs, and provide enough personal context for an agent to draft the post in that person’s writing style. These aren’t simply better chatbot responses. They are examples of software using accumulated knowledge about a person to decide what might be useful next.
This is where the idea gets particularly interesting. We have spent years interacting with computers largely on their terms. We open applications, find information, move it between services, and repeatedly explain what we want. Persistent context could reverse some of that relationship by allowing software to understand what is happening without requiring us to explain everything from scratch.
Qualcomm is already positioning Snapdragon around this broader idea of personal and agentic AI. At Snapdragon Summit, the company described AI that can work across PCs, phones, cars, earbuds, and other devices, with more processing happening locally. Its Snapdragon X Series strategy similarly emphasizes AI agents running directly on PCs, rather than depending entirely on remote servers.
Liquid Context could become one piece connecting those experiences. For example, context collected during a run on a smartwatch could later reach a car, allowing its agent to know about a new personal record, adjust the cabin based on the user’s preferences, and suggest stopping for a recovery drink. The important part isn’t the smoothie suggestion itself, but that context could potentially follow a user between different devices.
Liquid AI is also offering Liquid Agent, an embedded agent based on its LFM2.5-2.6B model. The company says it has optimized the model for Qualcomm’s Hexagon NPU, giving device manufacturers another option if they want to build an embedded or hybrid AI assistant alongside Liquid Context. Liquid AI has been pushing its LFM2.5 family around the idea of models capable of running across different types of hardware.
There is an important caveat here. This announcement does not mean every Snapdragon phone or PC is suddenly going to start building a detailed memory about its owner. Liquid AI and Qualcomm Technologies are providing technology that device manufacturers can potentially incorporate into their own products, and the companies say they are exploring additional collaboration.
Still, the direction is becoming clear. Faster CPUs and GPUs will continue to matter, but NPUs are giving device makers dedicated hardware capable of running AI continuously without leaning on the cloud for everything. Qualcomm is openly describing Snapdragon as a platform for the agentic age, and personal context could be what makes those agents substantially more capable.
The privacy implications deserve just as much attention as the convenience. Keeping personal context on the device is preferable to uploading an endless stream of private information to somebody else’s servers, but local storage alone doesn’t settle the issue. Users will need clear control over what gets remembered, how long it remains there, which agents can access it, what leaves the device, and whether deleting that memory actually deletes it everywhere.
An AI assistant that knows almost nothing about you can be frustratingly useless, while one that knows almost everything about you could be incredibly useful. Whether people are comfortable with that tradeoff may become one of the biggest questions of the personal AI era.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz
