OPPO and Academic Partners Launch MobileMem Benchmark for On-Device AI Memory
OPPO, in collaboration with OpenKG and academic teams from Zhejiang University, Tongji University, and Southeast University, introduced MobileMem, a benchmark designed to evaluate on-device long-term AI memory for mobile agents. The technical whitepaper is now public, and the dataset alongside evaluation tools will be open-sourced. As mobile AI evolves from basic query-answering chatbots to proactive personalized agents, retaining user habits and historical context on-device becomes critical. MobileMem establishes a standardized testing framework that allows industry developers to compare, evaluate, and iterate different on-device memory technologies. MobileMem evaluates long-horizon memory using realistic mobile usage trajectories containing interleaved user-assistant dialogues and app interaction images. OPPO intends to leverage these memory foundations for its upcoming AI-centric operating system, ColorOS 17.
## BACKGROUND
On-device AI memory enables personal language models to store and retrieve contextual user preferences directly on mobile hardware without sending sensitive personal data to cloud servers. An AI Operating System (AI OS) integrates these model capabilities and agent scheduling into the core system layer, allowing intelligent assistants to autonomously execute tasks on behalf of the user.