Intent-Driven Network CoPilot: Updating Cross-Protocol Configurations for Edge Service Guarantee
Abstract
There is a growing demand for rapidly updating networks to meet service requirements in edge scenarios. Issues such as dynamic network and service changes pose higher challenges for configuration management. Traditional configuration management in static networks has limitations in interactivity, customization, and performance assurance, making it difficult to adapt to rapid changes. Therefore, an automated solution capable of understanding service intents and dynamically updating network configurations is urgently needed. We propose Network CoPilot, an intent-driven model NLI2Conf that integrates graph and language models to automate configuration updates via natural language. Our approach designs multimodal interfaces to ingest raw network data (configurations, topology, performance metrics, link states) without specialized languages. NLI2Conf supports intents related to forwarding strategies and performance constraints. It also supports performance intents for specified service traffic, enabling fine-grained guarantees for edge services. The model updates configurations to align intents with device behaviors while ensuring consistency. We collect data through a network-realistic process and train NLI2Conf via masked pre-training and LoRA fine-tuning. Extensive experiments demonstrate that NLI2Conf accurately understands the intents involving multiple protocols (OSPF, BGP, QoS) and raw network input, achieving 91% intent-configuration consistency.