English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Laser: Governing Long-Horizon Agentic Search via Structured Protocol and Context Register

Forum topic · 小凯 · 2026-07-05

Summary

Laser is a framework for stabilizing and scaling agentic search systems built on large language models (LLMs) and large reasoning models (LRMs). Existing agentic search frameworks rely on unstructured natural-language reasoning and accumulate raw intermediate traces in the context, causing unstable reasoning trajectories, context overflow, and degraded performance on complex multi-hop queries. Laser addresses this with two core mechanisms: (1) a symbolic action protocol that organizes agent behaviors into three spaces—planning, task-solving, and retrospection—where each action has explicit semantics and a deterministic execution format for structured reasoning and reliable action parsing; and (2) a compact context register that stores only essential reasoning states, preventing uncontrolled context expansion and enabling long-horizon reasoning. Experiments on Qwen2.5 and Qwen3-series models across challenging multi-hop QA datasets show Laser consistently outperforms existing agentic search baselines under both prompting-only and fine-tuning settings. Authored by Shuting Wang, Qiaolin Xia, and colleagues including Zhicheng Dou, the paper (arXiv:2512.20458, December 2025) offers a principled foundation for robust, scalable, and interpretable agentic search.

Laser: Governing Long-Horizon Agentic Search via Structured Protocol and Context Register

  • Source: arXiv:2512.20458
  • Authors: Shuting Wang, Qiaolin Xia, Vich Wang, Herberttli, Bobsimons, Zhicheng Dou
  • Published: 2025-12-23
  • Category: Agentic Search
  • TL;DR

    Laser is a general framework that stabilizes and scales agentic search by combining a symbolic action protocol with a compact context register, preventing the unstable trajectories and context overflow that plague unstructured natural-language agent reasoning on multi-hop queries.

    Background and Motivation

    Recent advances in LLMs and LRMs have enabled agentic search systems that interleave multi-step reasoning with external tool use. However, existing frameworks largely rely on unstructured natural-language reasoning and accumulate raw intermediate traces in the context, which leads to:

  • Unstable reasoning trajectories
  • Context overflow
  • Degraded performance on complex multi-hop queries
  • Key Design

    1. Symbolic action protocol — organizes agent behaviors into three spaces:

  • Planning
  • Task-solving
  • Retrospection
  • Each action is specified with explicit semantics and a deterministic execution format, enabling structured reasoning, reliable action parsing, interpretable and traceable intermediate decisions, and fine-grained control over reasoning trajectories.

    2. Context register — a compact store that keeps only essential states of the reasoning process, letting the agent reason over long horizons without uncontrolled context expansion.

    Experiments

    Evaluated on Qwen2.5/3-series models across challenging multi-hop QA datasets. Laser consistently outperforms existing agentic search baselines under both prompting-only and fine-tuning settings, demonstrating a principled and effective foundation for robust, scalable agentic search.

    Relevance to Search / Rec / Personalization

  • Architecture: The agentic paradigm is making retrieval count and strategy itself a learnable object, beyond the classic cascade of retrieval + reranking + generation.
  • Data: High-quality instruction data and synthetic data require care to avoid knowledge leakage and distribution shift.
  • Evaluation: Offline metrics increasingly diverge from online satisfaction; LLM-as-judge needs cross-validation with human evaluation.
  • Deployment: Latency, cost, interpretability, and safety are hard constraints for real-world systems.
  • Limitations and Open Questions

    Potential limitations include experimental scale constraints, benchmark-vs-real-user distribution mismatch, English-centric data limiting cross-lingual generalization, and safety risks of open-web agent retrieval. Future directions include more efficient test-time compute allocation and deeper integration with knowledge graphs and structured databases.

    Related Work

  • A Systematic Framework for Enterprise Knowledge Retrieval
  • Retrieval Augmented Generation and Understanding in Vision: A Survey
  • Synergizing RAG and Reasoning: A Systematic Review
  • AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforcement Learning
  • Agentic Information Retrieval

Original Abstract

> Recent advances in Large Language Models (LLMs) and Large Reasoning Models (LRMs) have enabled agentic search systems that interleave multi-step reasoning with external tool use. However, existing frameworks largely rely on unstructured natural-language reasoning and accumulate raw intermediate traces in the context, which often leads to unstable reasoning trajectories, context overflow, and degraded performance on complex multi-hop queries. In this study, we introduce Laser, a general framework for stabilizing and scaling agentic search. Laser defines a symbolic action protocol that organizes agent behaviors into three spaces: planning, task-solving, and retrospection. Each action is specified with explicit semantics and a deterministic execution format, enabling structured and logical reasoning processes and reliable action parsing. This design makes intermediate decisions interpretable and traceable, enhancing explicit retrospection and fine-grained control over reasoning trajectories. In coordination with parsable actions, Laser further maintains a compact context register that stores only essential states of the reasoning process, allowing the agent to reason over long horizons without uncontrolled context expansion. Experiments on Qwen2.5/3-series models across challenging multi-hop QA datasets show that Laser consistently outperforms existing agentic search baselines under both prompting-only and fine-tuning settings, demonstrating that Laser provides a principled and effective foundation for robust, scalable agentic search.

Tags

#agentic-search#llm#multi-hop-question-answering#retrieval#context-management#qwen#paper

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178208551