Reproduce “XferAgent: Cross-Agent Experience Transfer for Self-Improving LLM Agents” 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

Weihao Li, Jun Bai, Shijie Chen, Qincheng Lu, Xinyu Wang, Ziyang Song

XferAgent transfers reusable success and failure lessons across agents by reconciling them in a shared graph, routing connected task-relevant lessons, and combining them with local memory. On 301 GAIA test tasks, GPT-5.6 Luna with XferAgent achieves 52.16% accuracy, versus 47.51% for ExpGraph and 48.17% with no memory. This benchmark-specific result does not show memory helps in every setting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.