Reproduce “Evolver: Turning Agent Exploration into Evolving Capabilities” 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
Haoyang Zhang, Qumeng Sun, Xiao Zhang, Xiang Liu, Zhidong Fan, Junjie Wang
Evolver lets existing agents reuse task-matched procedures extracted from earlier runs, without retraining the underlying model. It stores procedural guidance separately, then retrieves and injects relevant guidance into a later agent session. In a diagnostic paired evaluation prioritizing same-task experience, success rose from 83.13% to 87.50%, while recorded task-execution tokens fell 24.24%; extraction and other overhead were excluded.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.