Demo: A Real-time Multi-Agent Network Attack Detection and Incident Response System

Network attacks have disrupted critical infrastructure and compromised important network operations worldwide for several decades. In spite of tremendous research outcomes based on various artificial intelligence techniques, current solutions still struggle to detect unknown network attacks and to adapt pretrained models in real time. This demo paper proposes a real-time multi-agent system to detect and react to the network dynamically and efficiently by using the LLM and RAG techniques. The proposed system consists of multiple agents: a data processing agent for feature engineering, a detection agent for traffic pattern analysis and attack detection, and a response agent for selecting appropriate response actions in the network in real time. Finally, based on RAG techniques, it provides a recommendation report for network administrators by searching databases of well-known vulnerability repositories, such as CVE. We demonstrate that the proposed system detects real-time network data with high accuracy. Its processing time is efficient enough to handle large volumes of network traffic.