Tag
This paper introduces PerMemBench, the first benchmark for evaluating personalized memory systems in LLM-based agents, and proposes a session-level storage gating framework that adapts memory policies to individual user contexts.