Reproduce “SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration” 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
Yuyao Zhang, Junjie Gao, Zhengxian Wu, Jiaming Fan, Jin Zhang, Shihan Ma, Yao Yao, Weiran Qi and 6 more
SearchOS-V1 introduces a system-level multi-agent framework for robust open-domain information-seeking, externalizing search state into explicit, shared components and employing pipeline-parallel scheduling. The system achieved the best overall F1 scores on WideSearch, improving by 4.3 points, and reduced average end-to-end search time by 24.3% through enhanced coordination and error handling.
Key insights
- Externalizing and persistently managing search state (progress, evidence, failures) outside of conversational history is crucial for improving robustness and efficiency in long-horizon tasks.
- Adaptive, search-time schema planning, which dynamically selects the appropriate relational structure, consistently outperforms universally fixed schemas for open-domain information seeking.
- Pipeline-parallel scheduling and continuous task dispatch significantly improve resource utilization and throughput compared to traditional batch processing in multi-agent environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.