arXiv：cs.AI· Tairan Huang, Siyu Shang, Qiang Chen, Xiu Su, Yi Chen·· 16 小时前
FlowAgent：将工具调用变为连续流以演进智能体推理
Tools as Continuous Flow for Evolving Agentic Reasoning
AI 导读
针对现有 LLM 工具调用逐步式范式缺乏全局视角、长程易累积错误且难以泛化到未见工具的问题，研究者提出 FlowAgent，将工具链重构为语义空间内的连续轨迹生成，用条件流匹配生成连续隐轨迹以提供全局规划视角。作者还提出首个面向动态真实环境的 plan 级闭环基准，并从理论上给出效用收敛边界，证明连续形式可保证稳健泛化与误差衰减。该工作已被 NeurIPS 2026 接收。
正文
Abstract:Large Language Models (LLMs) have demonstrated remarkable capabilities in orchestrating tools for reasoning tasks. However, existing methods rely on a step-wise paradigm that lacks a global perspective, which causes error accumulation over long horizons and restricts generalization to unseen tools. To overcome these limitations, we propose Tools as Continuous Flow for Evolving Agentic Reasoning (FlowAgent), which reconceptualizes tool chaining as continuous trajectory generation within a semantic space. To systematically evaluate this paradigm, we introduce the first plan-level closed-loop benchmark dedicated to plan-level agentic reasoning in dynamic real-world environments. Specifically, the proposed FlowAgent leverages conditional flow matching to generate continuous latent trajectories, providing a global planning perspective to ensure coherent and robust tool execution. Theoretically, we establish formal bounds on utility convergence and prove that our continuous formulation fundamentally guarantees robust generalization and error attenuation. Empirical evaluations show that FlowAgent achieves superior robustness and adaptability in long-horizon reasoning tasks.
| Comments: | Accepted by NeurIPS 2026 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.07339 [cs.AI] |
| (or arXiv:2605.07339v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.07339 arXiv-issued DOI via DataCite |
Submission history
From: Tairan Huang [view email]
[v1]
Fri, 8 May 2026 06:44:11 UTC (5,240 KB)
[v2]
Wed, 12 Aug 2026 11:49:34 UTC (7,869 KB)
[v3]
Sat, 10 Oct 2026 02:51:54 UTC (7,874 KB)
来源：arXiv：cs.AI · arxiv.org