Reproduce “System Optimizations for Cost-Efficient Agent Harnesses” 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
Jongseok Park, Jiaxiang Yu, Eric Zhou, Tony Hong, Sunga Kim, Yusuf Afifi, Ian Wong, Ion Stoica and 1 more
Agency inserts an interposition layer between harnesses, models, and tools to reduce context and calls. Its techniques compact context, remove stale history, search metadata, and batch calls. Across 200 tasks, four models, and three harnesses, they average 0.77 of baseline cost and 0.97 of baseline accuracy, though some tasks cost more and accuracy declines on some OfficeQA tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.