Reproduce “Federation over Harness” 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

Dixi Yao, Tahseen Rabbani, Tian Li

FoH lets federated AI agents evolve both their task procedures and the server process that turns local reflections into shared updates. Human-written guides constrain changes, while reflections—not ground-truth labels or validation data—drive evolution. On ClawEval, FoH scores 78.44 versus 56.60 for isolated ReAct agents; the 23 PinchBench V1 tasks used for evolution are separate from evaluation.

Summary by alphaXiv.

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