Reproduce “Discovering Effective Agent Memory Designs through Program Evolution” 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

Xinheng Ding, Seyyed Saeid Cheshmi, Ziyang Zhang, Tairan Chen, Wenya Xie, Yu Hou, Rui Zhang, Zirui Liu

MAPE evolves conversational memory programs that build, retrieve, and compose evidence for answering questions about past chats. An LLM proposes program changes, and execution traces diagnose failures; the search can optimize accuracy alone or accuracy and estimated serving cost. On held-out LoCoMo with Qwen3-30B-A3B-Instruct-2507, the cost-aware system reaches 81.33% accuracy at 35% of SimpleMem’s serving cost, in this benchmark and backbone.

Summary by alphaXiv.

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