I benchmarked memory for AI agents, and plain files did just as well
Third-party memory engines, WireCat Memo and a plain folder of files on the same questions. What the agent answers, which record it cites and what it costs, plus the public LifeBench.

