MeshAgent: Enabling Reliable Network Management withLarge Language Models

Yajie Zhou

doi:10.1145/3771567

MeshAgent: Enabling Reliable Network Management withLarge Language Models

2025-12-01

0

PRE

AI

DOI：10.1145/3771567

Original request for help

Save

Abstract

En 中文

The emergence of large language models (LLMs) offers great promise for building domain-specific agents,but adapting them for network management remains challenging. To understand why, we conduct a casestudy on network management tasks and find that state-of-the-art specialization techniques rely heavily onextensive, high-quality task-specific data to produce precise solutions. However, real-world network queriesare often diverse and unpredictable, making such techniques difficult to scale. Motivated by this gap, wepropose MeshAgent1, a new workflow that improves precision by extracting domain-specific invariants fromsample queries and encoding them as constraints. These constraints guide LLM's generation and validationprocess, narrowing the search space and enabling low-effort adaptation. We evaluate our method across threenetwork management applications and a user study involving industrial network professionals, showingthat it complements existing techniques and consistently improves accuracy. We also introduce reliabilitymetrics and demonstrate that our system is more dependable, with the ability to abstain when confidenceis low. Overall, our results show that MeshAgent achieves over 95% accuracy, reaching 100% when pairedwith fine-tuned agents, and improves accuracy by up to 26% compared to baseline methods. The extraction ofreusable invariants provides a practical and scalable alternative to traditional LLM specialization, enabling thedevelopment of more reliable agents for real-world network management.

Keywords:

Large Language Models (LLMs) for Networking

Agentic System Reliability

MeshAgent: Enabling Reliable Network Management withLarge Language Models

Abstract

Journal

Organization