StreamMemBench: Streaming Evaluation of Agent Memory for Future-Oriented Assistance

In three linesStreamMemBench is an evaluation benchmark for testing AI agent memory in realistic scenarios. It constructs two-step task sequences around video observations (EgoLife) to measure whether agents use stored evidence and reuse user feedback. Tests on 8 memory systems show current agents often fail to convert feedback into reliable follow-up behavior.

Summary generated by Claude — human-verified