ARCHIVES

Enterprise configuration management has become increasingly challenging due to the growth of distributed systems, strict security requirements, and complex interdependencies among software components. Administrators are required to interpret extensive documentation while ensuring that configuration changes align with the actual system state. Although large language models have shown promise in providing natural language assistance, their direct application to configuration management raises concerns related to hallucination, lack of contextual grounding, and operational safety. This paper presents an agentic artificial intelligence framework that combines hierarchical retrieval-augmented generation, knowledge graph‒based system modelling, and human-in-the-loop controlled execution. The framework enables administrators to query configuration procedures in natural language, grounds responses in authoritative documentation, validates recommendations against live infrastructure topology, and executes actions only after explicit human approval. System-level evaluation demonstrates reduced task completion time, lower configuration error rates, and improved explain ability compared to manual and text-only AI approaches. The results indicate that agentic AI architectures are well suited for enterprise environments where reliability and accountability are critical. Traditional documentation-centric support systems and conventional AI chatbots remain largely passive, offering static guidance without sufficient grounding in real-time system state or operational safeguards. Addressing these limitations, this research presents Admin Assist AI agent, an intelligent administrative assistant that integrates structure-aware Retrieval-Augmented Generation (RAG), a stateful Knowledge Graph, and a Human-in-the-Loop (HITL) execution framework to deliver trustworthy, explainable, and safe operational assistance