Reproduce “MemoryTax: The Bitter Lesson of Agent Memory” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zedong Peng
The authors compare agent search of raw conversation records with prebuilt memory systems on one LoCoMo conversation. Agents use tools to find evidence, while Mem0 and Zep construct stores before questions arrive. With Claude Opus 5.5, Pi scores 99% on 152 questions at $6.95 per rollout, versus 95% for full context; photo-caption access and one run limit the comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.