MobileMem: Learning from a Year of Mobile Experiences
06:00 · August 17, 2026 · arXiv cs.AI RSS
The next generation of AI agents is increasingly moving beyond systems that answer isolated questions toward persistent personal assistants that can understand, remember, and continuously learn from users' experiences. Such assistants require long-term memory to accumulate and leverage user-specific experiences over time, yet existing benchmarks remain inadequate for realistic mobile settings, where experiences are heterogeneous, multimodal, evolving, and deeply personal. We introduce MobileMem, a benchmark and framework for studying on-device long-term memory, grounded in a year-scale collection of mobile experiences. MobileMem employs a knowledge-grounded synthesis pipeline to construct coherent and temporally consistent long-horizon trajectories from user-app sessions. It provides complementary text and multimodal settings covering multi-hop and temporal reasoning, knowledge updating, and implicit preference inference. Specifically, MobileMem enables agents to remember the past, understand the present, and adapt to the future. By modeling experiences rather than isolated facts, MobileMem moves memory beyond information retrieval toward experiential intelligence for continuous personal learning.
Summary
MobileMem introduces a benchmark and framework for evaluating long-term memory in on-device AI agents, built from a year-scale collection of realistic mobile user-app interactions. Rather than relying on isolated facts or static datasets, the system constructs coherent, temporally consistent trajectories that span multiple sessions, modalities, and applications. A knowledge-grounded synthesis pipeline generates these long-horizon sequences by anchoring synthetic experiences to structured prior knowledge, personas, and temporal constraints, producing both text-only and multimodal test cases.
The benchmark targets capabilities that current cloud-oriented memory systems rarely address. It assesses multi-hop and temporal reasoning over evolving user contexts, the ability to update stored knowledge as preferences and routines change, and the inference of implicit user preferences from fragmented event streams. By modeling experiences instead of discrete facts, MobileMem shifts evaluation from simple retrieval toward what the authors term experiential intelligence, enabling agents to remember past events, interpret present situations, and adapt future behavior without catastrophic forgetting.
The work underscores the practical constraints of mobile deployment. On-device memory must operate under tight limits on storage, power, and compute while handling continuous multimodal data from chat logs, photos, screenshots, and location traces. It also avoids the privacy risks of uploading personal histories to centralized servers. MobileMem therefore frames on-device memory as a distinct research direction, one that requires new architectures for organizing, compressing, and updating personal knowledge graphs locally over extended periods.
Why it matters
This research is highly relevant for Dutch AI researchers and developers focusing on edge AI and personal assistants. Its emphasis on on-device, local-first memory processing aligns perfectly with the EU's strict GDPR privacy standards, offering a practical framework for building compliant, personalized AI systems.
