Empowering network management with a multi-agent LLM architecture

A novel, to our knowledge, multi-agent architecture powered by open-source large language models (LLMs) is proposed for managing optical network control, designed to meet the growing demand for flexible and scalable automation. The system dynamically retrieves device-specific client code and task instructions while ensuring full compliance with confidentiality requirements by integrating an on-premises content database. Its modular design accommodates various open-source LLM frameworks, allowing different models to be seamlessly deployed across diverse network scenarios. This approach eliminates reliance on cloud services, safeguarding sensitive data while enabling automated provisioning, monitoring, and inventory updates. The extensive device pool covers a wide range of hardware platforms, validated through comprehensive multi-vendor testing. Additionally, the framework’s adaptability is demonstrated by evaluating performance across varying user expertise levels, from novice operators to advanced professionals. Experimental outcomes indicate consistent reliability and high accuracy, irrespective of operator skill. Overall, the proposed architecture enables LLM-driven, modular systems that streamline complex network operations, preserve data privacy, and support interoperability across vendors, presenting a forward-looking solution for large-scale device management.