Integrating LLMs with NetBox and Netmiko for Vendor-Agnostic Intent-Based Networking
Lucas Immanuel Nickel, Lorenz Hohmann, Nick Stolbov, Lukas Gerstacker, Sebastian Rieger · 2025
The operation of modern network infrastructures presents network operations engineers with considerable challenges, esp. in multi-vendor environments, due to their inhomogeneity and technological diversity. Intent-based networking (IBN) offers a solution to reduce this complexity by allowing users to specify desired network states as intents, which are then translated into network policies and continuously monitored. This paper presents an approach for the translation and implementation of intents based on Retrieval-Augmented Generation (RAG) that is integrated with the network automation tools Netmiko and NetBox. Intents can be used to express and perform changes in these tools, e.g., to automate otherwise tedious and hence error-prone tasks to assist human network administrators. The evaluation shows a 90.2% success rate for individual phases (Intent Conception, Request Construction, Response Correctness) and 81.2% end-to-end. To safeguard network availability and integrity, the Large Language Model (LLM) references manufacturer documentation online and prompts operators for confirmation. The implementation is provided as open source. Potential network management and operations use cases benefiting from LLMs and RAG are discussed based on the accuracy, correctness, and practicability of the presented testbed.