1 paper · 1 filter
Bowen Qin, Yao Lu
For a long-horizon LLM agent, the memory question is not what was once recorded but what \emph{currently holds}. Most designs answer it only indirectly: every interaction is stored…