Reproduce “Multi-Agent Transactive 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

To Eun Kim, Xuhong He, Dishank Jain, Ambuj Agrawal, Negar Arabzadeh, Fernando Diaz

MATM lets heterogeneous agents share successful action–observation trajectories, retrieving relevant segments for a consumer’s current state. A planner requests retrieval selectively, and a learned reranker estimates whether a candidate improves outcomes over no retrieval. Across 34 consumer models and five allocation runs on ALFWorld, SVMRank raises success from 0.4708 to 0.6431 and lowers steps; evidence is limited to two benchmarks.

Summary by alphaXiv.

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