InfoSeeker: A Scalable Hierarchical Parallel Agent Framework for Web Information Seeking

2026-04-03

AI Summary

This work addresses the challenges of context saturation, error cascading, and high latency faced by large language model agents in large-scale web information gathering. To overcome these issues, the authors propose a hierarchical parallel agent framework grounded in the principle of approximate decomposability, comprising a strategic Host layer, a Manager layer, and a set of parallel Worker agents. The Manager layer mitigates error propagation through context isolation, evidence aggregation, and reflective reasoning, thereby ensuring high-quality information synthesis, while the Worker layer enables task-level parallelism for accelerated execution. Experimental results demonstrate that the proposed approach achieves an 8.4% success rate on WideSearch-en and 52.9% accuracy on BrowseComp-zh, with a 3–5× speedup in overall execution time, substantially improving system scalability and efficiency.

Abstract

Recent agentic search systems have made substantial progress by emphasising deep, multi-step reasoning. However, this focus often overlooks the challenges of wide-scale information synthesis, where agents must aggregate large volumes of heterogeneous evidence across many sources. As a result, most existing large language model agent systems face severe limitations in data-intensive settings, including context saturation, cascading error propagation, and high end-to-end latency. To address these challenges, we present \framework, a hierarchical framework based on principle of near-decomposability, containing a strategic \textit{Host}, multiple \textit{Managers} and parallel \textit{Workers}. By leveraging aggregation and reflection mechanisms at the Manager layer, our framework enforces strict context isolation to prevent saturation and error propagation. Simultaneously, the parallelism in worker layer accelerates the speed of overall task execution, mitigating the significant latency. Our evaluation on two complementary benchmarks demonstrates both efficiency ($ 3-5 \times$ speed-up) and effectiveness, achieving a $8.4\%$ success rate on WideSearch-en and $52.9\%$ accuracy on BrowseComp-zh. The code is released at https://github.com/agent-on-the-fly/InfoSeeker

Problem

• information synthesis
• context saturation
• error propagation
• latency
• web information seeking

Innovation

• hierarchical agent framework
• parallel information seeking
• context isolation
• error propagation mitigation
• scalable web search