Reproduce “Inventa: Decentralized Self-Improving Agents for Autonomous Scientific Discovery” 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

Prithwish Jana, Nikos Kanakaris, Purak Jain, Anoop Deoras, Sahika Genc

The authors introduce Inventa, which proposes research ideas through adaptive parallel units and shares only verifier-approved findings. On 736 IdeaBench papers, it yields 26.0 distinct ideas rated novel by a judge per paper, versus 18.6 for ResearchAgent, using 45 versus 247 model calls. On EurekaBench, its conditioned score is 73.6% versus CORAL’s 56.2%, with no constraint failures in 16 tasks.

Summary by alphaXiv.

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