Network CoPilot: Intent-Driven Network Configuration Updating for Service Guarantee

Updating network configurations not only requires satisfying new forwarding strategies, but also faces critical performance constraints brought on by various services. However, existing methods face significant limitations in interaction form, customization, and performance guarantees, making it challenging to keep pace with the rapid development of enterprise networks. Therefore, we propose a network copilot, a natural language intent-driven model for network configuration updating called NLI2Conj, which integrates graph and language models. We design network multimodal interfaces based on text attribute graphs to input network information in its raw format, including configuration files, topology, and performance metrics. This eliminates the need for additional specialized language design and learning. NLI2Conf understands forwarding strategies and performance constraints in natural language, and updates existing configurations, achieving alignment between intents and device behaviors. We provide a configuration collection process that mirrors the real network, and collect data samples from strategy intent to network configuration to facilitate NLI2Conj's capability of configuration updating. Extensive experiments demonstrate that NLI2Conf can accurately interpret network information and strategy intent in raw format. The consistency between configuration and strategy intent reaches 96.6 %, and the proposal reduces labor costs to 1/26.