A new multi-agent network support chatbot, developed as a Master's thesis by en-medina, leverages LLM technology to automate enterprise IT help. Built using a LoRA-tuned LLaMA model and LangGraph, this system acts as a first line of defense, classifying issues and routing them to specialized agents. The architecture features a Triage Agent that achieves over 90% classification accuracy, directing requests to appropriate handlers like the Knowledge Agent.

The Knowledge Agent utilizes Pinecone for vector storage and Maximum Marginal Relevance (MMR) to query internal documentation, ensuring context-aware responses via RAG. For connectivity diagnostics, the Connectivity Agent employs a ReAct pattern to check IP reachability and DNS records. The Device Agent, built on Netmiko, uses a Plan-and-Execute strategy to interact directly with network hardware like Cisco routers, performing read and limited write operations.

Finally, the Escalation Agent manages sensitive or complex requests, automatically creating tickets in ClickUp. This solution addresses the growing complexity of modern networks, offering faster, smarter support while reducing the burden on human IT staff.