Reproduce “Evolving Agent Interfaces” 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
Saaket Agashe, Yuzhe Yang, Gonzalo Gonzalez-Pumariega, Bhuvan Chandra Koduru, Chih-Lun Li, Jiachen Yang, Ang Li, Xin Eric Wang
Agent S4 evolves repositories of computer-use tools and guidance by having a task agent attempt synthetic tasks and a meta-agent revise interfaces from the trajectories. A separate controller selects repositories using validation performance hidden from the meta-agent. On OSWorld V2, Qwen3.8-27B reaches 21.30% binary success, but comparison with fixed Qwen CUA changes harnesses and does not isolate repository effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.