LongSeeker: Elastic Context Orchestration for Long-Horizon Search Agents

• Authors: Yijun Lu, Rui Ye, Yuwen Du, Jiajun Wang, Songhua Liu, Siheng Chen
• Published: 2026-05-06
• Source: https://arxiv.org/abs/2605.05191
• Type: Academic paper | Section: Agentic Search

Background

Agentic search in large-scale retrieval, recommendation, and personalization systems has long faced challenges in efficiency, scalability, and intent understanding. Traditional pipelines separate retrieval, ranking, and generation, which struggles to meet LLM-era demands for natural-language interaction, multi-hop reasoning, and up-to-date knowledge. LongSeeker targets this intersection, focusing on how long-horizon search agents should manage a rapidly growing working context.

Core Contributions

• A unified view of adaptive context management for long-horizon agents, bringing scattered related work into a comparable framework.
• Context-ReAct, a general agentic paradigm integrating reasoning, context management, and tool use in a single loop, with five atomic operations: Skip, Compress, Rollback, Snippet, Delete.
• Theoretical guarantees: the Compress operator is expressively complete, while the other specialized operators provide efficiency and fidelity guarantees that reduce generation cost and hallucination risk.
• LongSeeker, a long-horizon search agent fine-tuned from Qwen3-30B-A3B on 10k synthesized trajectories, evaluated on four representative search benchmarks.
• Identification of open problems: evaluation trustworthiness, latency and cost, hallucination and safety, cross-lingual and multimodal extension.

Method

The approach follows: problem formalization → model/system design → training → inference pipeline. Long-horizon search agents must manage a rapidly growing working context as they reason, call tools, and observe information; naively accumulating all intermediate content overwhelms the agent. The key principle is that context management should be adaptive: parts of the agent's trajectory are maintained at different levels of detail depending on their current relevance to the task. The five atomic operations let the agent dynamically reshape its working context—preserving important evidence, summarizing resolved information, discarding unhelpful branches, and controlling context size.

Results

Across four representative search benchmarks, LongSeeker achieves:

• 61.5% on BrowseComp and 62.5% on BrowseComp-ZH
• Substantially outperforming Tongyi DeepResearch (43.2% and 46.7%) and AgentFold (36.2% and 47.3%)

These results highlight that adaptive context management allows agents to achieve more reliable and efficient long-horizon reasoning by actively shaping their working memory.

Insights for Search / Rec / Personalization

1. Architecture: cascaded retrieval + rerank + generation remains mainstream, but agentic paradigms make retrieval frequency and strategy themselves learnable. 2. Data: high-quality instruction data and click/session logs remain critical; synthetic data must guard against leakage and distribution shift. 3. Evaluation: offline metrics and online satisfaction are diverging; LLM-as-judge should be cross-validated with human evaluation. 4. Products: latency, cost, explainability, and safety are hard constraints for deployment—do not optimize academic benchmarks alone.

Limitations & Future Work

Potential limitations include experiment scale constrained by GPU budget, benchmarks mismatching real user distributions, English-centric data limiting cross-lingual generalization, and safety risks of agents operating on the open web. Future directions: more efficient test-time compute allocation, deeper integration with knowledge graphs and structured databases, and causal/fairness constraints for recommendation systems.

Original Abstract

> Long-horizon search agents must manage a rapidly growing working context as they reason, call tools, and observe information. Naively accumulating all intermediate content can overwhelm the agent, increasing costs and the risk of errors. We propose that effective context management should be adaptive: parts of the agent's trajectory are maintained at different levels of detail depending on their current relevance to the task. To operationalize this principle, we introduce Context-ReAct, a general agentic paradigm for elastic context orchestration that integrates reasoning, context management, and tool use in a unified loop. Context-ReAct provides five atomic operations: Skip, Compress, Rollback, Snippet and Delete, which allow the agent to dynamically reshape its working context, preserving important evidence, summarizing resolved information, discarding unhelpful branches, and controlling context size. We prove that the Compress operator is expressively complete, while the other specialized operators provide efficiency and fidelity guarantees that reduce generation cost and hallucination risk. Building on this paradigm, we develop LongSeeker, a long-horizon search agent fine-tuned from Qwen3-30B-A3B on 10k synthesized trajectories. Across four representative search benchmarks, LongSeeker achieves 61.5% on BrowseComp and 62.5% on BrowseComp-ZH, substantially outperforming Tongyi DeepResearch (43.2% and 46.7%) and AgentFold (36.2% and 47.3%). These results highlight the potential of adaptive context management, showing that agents can achieve more reliable and efficient long-horizon reasoning by actively shaping their working memory.