LongSeeker: Elastic Context Orchestration for Long-Horizon Search Agents

2026-05-06

AI Summary

This work addresses the challenges of cost and error propagation in long-horizon agents caused by context inflation during reasoning and tool use. The authors propose an adaptive context management mechanism, introducing the novel Context-ReAct paradigm that employs five atomic operations—Skip, Compress, Rollback, Snippet, and Delete—to enable flexible context orchestration. They theoretically establish the expressive completeness of the Compress operator while balancing computational efficiency and fidelity. Building upon Qwen3-30B-A3B fine-tuned with tens of thousands of synthetic trajectories, they develop a unified architecture integrating reasoning, context management, and tool invocation. Their approach achieves state-of-the-art performance with 61.5% and 62.5% accuracy on BrowseComp and BrowseComp-ZH benchmarks, respectively, significantly outperforming Tongyi DeepResearch and AgentFold.

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.

Problem

• long-horizon search
• context management
• working memory
• agent reasoning
• context growth

Innovation

• elastic context orchestration
• Context-ReAct
• long-horizon reasoning
• adaptive context management
• atomic context operations