Reproduce “AutoSwarm: Automating Agent Swarms” 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

Zhonghang Li, Zongwei Li, Chao Huang

AutoSwarm organizes agent contributions as scoped deliverables with acceptance conditions, evidence, and controlled handoffs tracked in shared execution state. On Plancraft-small, it achieved 74.36% overall success across 117 problems, versus Teamwork’s 63.25% in single-run evaluations. Removing supervisory review reduced success to 61.54%, supporting review’s importance in this setup, though the ablation does not isolate review alone.

Summary by alphaXiv.

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