Librarian Agents: Training LLMs to Curate Long-Context Workspaces that Anticipate Future Queries

Hugging Face Daily Papers Papers
long-context reinforcement-learning agentic-workflows retrieval-alternatives llm-reasoning workspace-curation

Summary

该论文提出 Librarian Agents 框架，利用强化训练 builder agent 将原始文档重组为可复用的层级化工作区，以服务未来的多样化查询。在 BrowseComp-Plus 上比最强检索与预计算基线高出 33 个百分点，并在 BRIGHT 上零样本泛化提升 31 个百分点。

Reasoning over long document contexts remains a challenge for large language models (LLMs). Standard retrieval methods are inherently reactive, bottlenecking agents with complex, cross-document reasoning at inference time, while precomputed structures like knowledge graphs often fall short because they rely on rigid heuristics. Instead, can we allow agents to perform anticipatory directed curation: learning to proactively reorganize unstructured corpora prior to inference to efficiently serve diverse future queries. We build on a key asymmetry in modern agentic harnesses (e.g., Claude Code): while LLM agents are adept at navigating existing filesystems to solve an immediate task, they are rarely trained to construct reusable workspaces for future use. To instantiate this capability, we introduce Librarian Agents, a framework in which a builder agent transforms raw documents into a hierarchical filesystem of synthetic files and reference links, which forms a workspace tailored for a downstream answering agent. We formulate workspace construction as a Markov decision process and optimize the builder policy via reinforcement learning to directly maximize downstream answering accuracy. Empirically, Librarian Agents outperforms the strongest retrieval and precomputation baselines by 33 percentage points on BrowseComp-Plus, amortizes upfront construction costs across repeated queries, and generalizes zero-shot to unseen domains on BRIGHT with a 31-point average accuracy improvement.
Original Article
View Cached Full Text

Cached at: 10/19/26, 07:30 PM

Paper page - Librarian Agents: Training LLMs to Curate Long-Context Workspaces that Anticipate Future Queries

Source: https://huggingface.co/papers/2610.22273

Abstract

Librarian Agents use reinforcement learning to train builder agents that reorganize document corpora into reusable hierarchical workspaces, improving downstream reasoning accuracy and amortizing costs across queries.

Reasoning over long document contexts remains a challenge forlarge language models(LLMs). Standardretrieval methodsare inherently reactive, bottlenecking agents with complex, cross-document reasoning at inference time, while precomputed structures likeknowledge graphsoften fall short because they rely on rigid heuristics. Instead, can we allow agents to perform anticipatory directed curation: learning to proactively reorganize unstructured corpora prior to inference to efficiently serve diverse future queries. We build on a key asymmetry in modernagentic harnesses(e.g., Claude Code): while LLM agents are adept at navigating existing filesystems to solve an immediate task, they are rarely trained to construct reusable workspaces for future use. To instantiate this capability, we introduceLibrarian Agents, a framework in which abuilder agenttransforms raw documents into ahierarchical filesystemof synthetic files and reference links, which forms a workspace tailored for a downstream answering agent. We formulate workspace construction as aMarkov decision processand optimize the builder policy viareinforcement learningto directly maximize downstream answering accuracy. Empirically,Librarian Agentsoutperforms the strongest retrieval and precomputation baselines by 33 percentage points onBrowseComp-Plus, amortizes upfront construction costs across repeated queries, and generalizes zero-shot to unseen domains onBRIGHTwith a 31-point average accuracy improvement.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2610\.22273

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.22273 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.22273 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.22273 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Let Agents Design How They Search: Automatic Interface Optimization for Agentic Search

arXiv cs.CL

论文提出 SearchWeaver 框架，利用代理的执行轨迹和任务奖励自动联合优化语料表示、可执行工具与使用指令，从而改进 agentic search 的搜索接口设计。使用 GPT-5.4-nano 时，优化后的接口在 BrowseComp-Plus 上将答案准确率从 71.6% 提升至 78.4%，并在四个 BRIGHT 子集上将平均 nDCG@10 从 45.7 提升至 55.0。

Libra: Training the Environment for Agentic Information Retrieval

arXiv cs.AI

The paper presents Libra, a self-evolving framework that introduces mutable catalogs into repositories to improve code localization for LLM agents, achieving logarithmic improvements and zero-shot transfer across different models and problem sets.

Training Advisors for LLM Agents from Task Outcomes

arXiv cs.CL

Nebius AI 提出 Caddie 方法，通过强化学习从任务成败结果训练 LLM agent 的自然语言批评器（critic），无需逐步标注。Qwen3-4B critic 在 MuSiQue 上将成功率提升逾 25 个百分点，并可跨基座模型与任务域泛化。