WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Xiaoshuai Song, Liancheng Zhang, Kangzhi Zhao, Yutao Zhu, Zhongyuan Wang, Guanting Dong, Jinghan Yang, Han Li, Kun Gai, Ji-Rong Wen, Zhicheng Dou
• Venue: arXiv: cs.CL
• Published: 2026-07-09
• Collected: 2026-07-10
• URL: http://arxiv.org/abs/2607.08662v1
• PDF: https://arxiv.org/pdf/2607.08662v1
• DOI: UNKNOWN

Plain-Language Brief

The paper introduces a new framework called WebSwarm, which uses multiple language model (LLM)-based search agents to handle complex, deep, and wide web searches more effectively than single agents or existing multi-agent systems. The system works by recursively delegating tasks and collaborating between agents, with each agent having its own local objective and search mode for organizing the search process.

What the Document Says

• The paper discusses a new framework called WebSwarm for deep-and-wide web search using LLM-based agents.
• A single ReAct-style agent is limited in handling depth and coverage simultaneously due to one long trajectory and limited context.
• Existing multi-agent systems have limitations in recursive depth, collaboration adaptability, and evidence-grounded expansion.
• WebSwarm constructs task decomposition, recursive expansion, and agent collaboration during inference.
• WebSwarm dynamically instantiates search nodes with local objectives and search modes for organizing search and collaboration.
• Each node can solve its objective or delegate child nodes. After solving, it returns evidence and results to parent nodes.
• WebSwarm probes task-relevant information organization on the web to guide subsequent node expansion.
• WebSwarm reuses process-level experience across homogeneous sibling nodes.
• Experiments show that WebSwarm outperforms single-agent and multi-agent baselines on deep, wide, and interleaved tasks.
• Further analyses provide insights for improving multi-agent search systems.

Key Concepts Mentioned

• Large language model (LLM)
• WebSwarm
• ReAct-style agent
• Task decomposition
• Recursive expansion
• Agent collaboration
• Search nodes
• Local objectives
• Search modes
• BrowseComp-Plus
• WideSearch
• DeepWideSearch
• GISA

Claims Made by the Document

• WebSwarm consistently outperforms single-agent and multi-agent baselines on deep, wide, and interleaved tasks.
• WebSwarm's effectiveness is explained by ablation, task difficulty, web tool efficiency, and model generalization.

Evidence Used

• Results from experiments on BrowseComp-Plus, WideSearch, DeepWideSearch, and GISA.

Methods / Approach

• Recursive delegation framework
• Dynamically instantiating search nodes
• Probing web information organization

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The provided text lacks information about potential limitations of the proposed method as stated by the authors.

Open Questions

• What are the specific algorithms or techniques used for task decomposition and recursive expansion?
• How does WebSwarm handle potential conflicts or redundancies between search nodes?

Notes for Future Q&A

May include interpretation. Labeled where it does.

• The paper focuses on the benefits and effectiveness of WebSwarm, but does not provide detailed information about its inner workings.

Raw Abstract / Excerpt

Large language model (LLM)-based web search agents are transforming information seeking from simple factoid question answering into complex, deep-and-wide search and research-oriented tasks. A single ReAct-style agent is constrained by one long trajectory and limited context, making it difficult to handle depth and coverage simultaneously. Existing multi-agent systems improve search coverage through parallel execution and aggregation, but still exhibit clear limitations in recursive depth, collaboration adaptability, and evidence-grounded expansion. We propose WebSwarm, a progressive recursive delegation framework that jointly constructs task decomposition, recursive expansion, and agent collaboration during inference. WebSwarm dynamically instantiates agentic search nodes, each coupling a local objective with a search mode that specifies how the node should organize search and collaboration. Each node can either solve its objective itself or further delegate child nodes; after solving, it returns evidence and results upward, enabling parent nodes to further expand, revise, or aggregate the search process. To guide this process, WebSwarm first probes how task-relevant information is organized on the web to ground subsequent node expansion, and reuses process-level experience across homogeneous sibling nodes. Experiments on BrowseComp-Plus, WideSearch, DeepWideSearch, and GISA show that WebSwarm consistently outperforms single-agent and multi-agent baselines on deep, wide, and interleaved deep-and-wide tasks. Further analyses of ablation, task difficulty, web tool efficiency, and model generalization explain WebSwarm's effectiveness and provide insights for multi-agent search systems.