Reproduce “WASP-Mem: Let Agent Organize Procedural Memory. Human Spend Effort on Diagnosis” 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
Peizhi Niu, Ge Shi, Vaibhav Gorde, Sarthak Ghosh, Christos Faloutsos, Min Cheng, Kevin Chen
WASP-Mem builds a procedural wiki of reusable action guidance from task attempts and feedback. The agent retrieves pages, executes tasks, and consolidates traces into updates; human diagnosis helps interpret failures. On TheAgentCompany, GPT-5.4 reaches a 72.5% checkpoint pass rate with a 233 KB store, versus 67.9% and 592 KB for Schema Form, in a one-round comparison that does not measure end-to-end task completion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.