AgentKVShift: cutting the cost of re-reading agent memory by up to 3.5x

AgentKVShift speeds up memory-augmented AI agents by cheaply correcting stale cache entries instead of fully re-encoding retrieved memories every time, the authors report.