SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration
Ant Group / Renmin University
Proposes SearchOS, a multi-agent framework for web-based information-seeking agents that converts fragile, implicit search progress into explicit, persistent, shared state via schema-completion-with-citations, a search-oriented context manager tracking task frontiers and evidence graphs, pipeline-parallel scheduling, and a hierarchical skill system. Shows measurable gains over single- and multi-agent baselines on benchmark datasets.
Why it matters
Top-voted paper on Hugging Face Daily Papers for 2026-07-17 with 35 upvotes; addresses a real bottleneck (state fragmentation) in long-horizon agent collaboration, relevant to anyone building research/search agents.
Importance: 3/5
Top-voted HF Daily Paper (35 upvotes) addressing multi-agent state fragmentation.