Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FS-Researcher: Test-Time Scaling for Long-Horizon Research Tasks with File-System-Based Agents

Chiwei Zhu, Benfeng Xu, Mingxuan Du, Shaohan Wang, Xiaorui Wang, Zhendong Mao, Yongdong Zhang

TL;DR

FS-Researcher tackles the challenge of long-horizon deep research tasks that exceed model context windows by introducing a persistent, file-system-based workspace. It splits labor into a Context Builder that crawls the web and curates a hierarchical knowledge base, and a Report Writer that assembles the final report on-demand from the KB, enabling iterative refinement across sessions. The approach achieves state-of-the-art report quality on DeepResearch Bench and DeepConsult across multiple backbones, with a clear positive link between context-building effort and final quality, demonstrating effective test-time scaling. The work highlights practical benefits for long-form scientific and analytical reporting, and provides open-source code and data for reproducibility and further development.

Abstract

Deep research is emerging as a representative long-horizon task for large language model (LLM) agents. However, long trajectories in deep research often exceed model context limits, compressing token budgets for both evidence collection and report writing, and preventing effective test-time scaling. We introduce FS-Researcher, a file-system-based, dual-agent framework that scales deep research beyond the context window via a persistent workspace. Specifically, a Context Builder agent acts as a librarian which browses the internet, writes structured notes, and archives raw sources into a hierarchical knowledge base that can grow far beyond context length. A Report Writer agent then composes the final report section by section, treating the knowledge base as the source of facts. In this framework, the file system serves as a durable external memory and a shared coordination medium across agents and sessions, enabling iterative refinement beyond the context window. Experiments on two open-ended benchmarks (DeepResearch Bench and DeepConsult) show that FS-Researcher achieves state-of-the-art report quality across different backbone models. Further analyses demonstrate a positive correlation between final report quality and the computation allocated to the Context Builder, validating effective test-time scaling under the file-system paradigm. The code and data are anonymously open-sourced at https://github.com/Ignoramus0817/FS-Researcher.

FS-Researcher: Test-Time Scaling for Long-Horizon Research Tasks with File-System-Based Agents

TL;DR

FS-Researcher tackles the challenge of long-horizon deep research tasks that exceed model context windows by introducing a persistent, file-system-based workspace. It splits labor into a Context Builder that crawls the web and curates a hierarchical knowledge base, and a Report Writer that assembles the final report on-demand from the KB, enabling iterative refinement across sessions. The approach achieves state-of-the-art report quality on DeepResearch Bench and DeepConsult across multiple backbones, with a clear positive link between context-building effort and final quality, demonstrating effective test-time scaling. The work highlights practical benefits for long-form scientific and analytical reporting, and provides open-source code and data for reproducibility and further development.

Abstract

Deep research is emerging as a representative long-horizon task for large language model (LLM) agents. However, long trajectories in deep research often exceed model context limits, compressing token budgets for both evidence collection and report writing, and preventing effective test-time scaling. We introduce FS-Researcher, a file-system-based, dual-agent framework that scales deep research beyond the context window via a persistent workspace. Specifically, a Context Builder agent acts as a librarian which browses the internet, writes structured notes, and archives raw sources into a hierarchical knowledge base that can grow far beyond context length. A Report Writer agent then composes the final report section by section, treating the knowledge base as the source of facts. In this framework, the file system serves as a durable external memory and a shared coordination medium across agents and sessions, enabling iterative refinement beyond the context window. Experiments on two open-ended benchmarks (DeepResearch Bench and DeepConsult) show that FS-Researcher achieves state-of-the-art report quality across different backbone models. Further analyses demonstrate a positive correlation between final report quality and the computation allocated to the Context Builder, validating effective test-time scaling under the file-system paradigm. The code and data are anonymously open-sourced at https://github.com/Ignoramus0817/FS-Researcher.
Paper Structure (37 sections, 2 equations, 6 figures, 9 tables)

This paper contains 37 sections, 2 equations, 6 figures, 9 tables.

Table of Contents

  1. Introduction
  2. FS-Researcher
  3. Architecture
  4. Tools.
  5. Workflow.
  6. Workspace.
  7. Context Builder
  8. Report Writer
  9. Experiments
  10. Experimental Setups
  11. Benchmarks.
  12. Baselines.
  13. Main Results
  14. Analyses
  15. Scaling Effect of Context Builder
  16. ...and 22 more sections

Figures (6)

  • Figure 1: Different deep research paradigms: (1) Top: Static pipelines and naive single agents that put raw observations in the context; (2) Middle: Agents whose trajectories are extended by compressing the observations, while still bounded by the hard context limit; (3) Bottom: FS-Researcher, an agent framework built on top of an external file system workspace with unlimited context size.
  • Figure 2: The framework of FS-Researcher.
  • Figure 3: Knowledge base example.
  • Figure 4: Left: KB statistics under 3-10 rounds of context-building. The number of characters in report corresponds to the y-axis on the right. Right: DeepResearch Bench scores of FS-Researcher with 3-10 rounds of context building.
  • Figure 5: Tool usage heatmap for the Context Building stage (first three iterations).
  • ...and 1 more figures