Announced at Snapdragon Summit 2026, Liquid AI, with its new partnership, announced that it has optimized its on-device context layer Liquid Context technology for Snapdragon processors. Leveraging Qualcomm’s Hexagon NPU hardware power, the technology processes user information directly on the device without sending it to the cloud, paving the way for AI spies to work in a much more individual, proactive, and privacy-focused way.

AI Creating Personal Memory Without Going to the Cloud

For AI assistants to truly help the user, it depends on how well they understand the user’s habits, priorities, and daily routines. This is exactly where Liquid Context comes in. The system, which constantly analyzes on-device signals (calendar, notifications, position, or sensor data) in accordance with the user’s permissions, builds a local memory layer.

This layer acts as a reliable memory function that can be used by all AI models, whether independent spies running on the device or third-party cloud services. The biggest advantage is that this analysis takes place entirely on the device; that is, no information is transferred to the cloud for each data update, thus reducing latency and protecting user privacy.

What’s Changing in Daily Life?

• Emergency Management: For example, when you receive a message from school that your child needs to be picked up early, the system scans your calendar to identify meetings that can be postponed and, with your approval, prepares draft planning emails.
• Event and End-of-Day Summaries: Following a busy conference or workday, the system filters the context of the day and your photo archive to create a draft LinkedIn or social media post that suits your style.
• Seamless Cross-Device Context: Information from a run you take with your smartwatch without your phone is transferred to the in-car system when you get into your car; While the vehicle adjusts the climate control according to your personal preference, it also suggests a route for a roadside beverage break.

High-Level Efficiency with Hexagon NPU and LFM2.5-2.6B Model

An optimized hardware architecture is critical to prevent a constantly running AI model from draining battery life. Liquid AI has optimized the LFM2.5-2.6B model, which feeds both the Liquid Context memory layer and its own embedded AI spy platform, Liquid Agent, specifically for Qualcomm’s Hexagon NPU unit. This structure, which combines scalar, vector, and tensor calculations with hardware accelerators, provides continuous learning with minimal power consumption. Hardware manufacturers (OEMs) can integrate the Liquid Agent model into their own devices or connect their own developed AI assistants directly to the Liquid Context layer thanks to this infrastructure. This development stands out as a valuable step in the transformation of artificial intelligence from a passive chatbot into a proactive assistant architecture that can anticipate user needs in advance.

Liquid AI and Qualcomm announced the Liquid Context layer, which offers on-device memory for Snapdragon processors. Details in our news!

Liquid, Artificial Intelligence, Data, Device, Day