Configuring and maintaining an IoT system is often a challenging task for non-technical users, particularly when troubleshooting becomes necessary on certain devices. A setup service function using AI models should be implemented to assist users, where the guidance offered is dynamically updated for real-time troubleshooting, as technology is constantly evolving. Retrieval-Augmented Generation (RAG) provides a unique solution to this challenge by combining information retrieval from external document data with Generative AI. RAG enables the system to retrieve real-time, relevant knowledge from external sources and provide accurate, context-aware advice without the need for constant model retraining or fine-tuning. In this study, we present an implementation of a RAG-based real-time troubleshooting capability within the setup service function for the SEMAR IoT application server platform. By leveraging data in SEMAR, the system dynamically pulls up-to-date information related to device issues, performance metrics, and known solutions in real time. This dynamic feature allows the system to handle emerging issues and troubleshooting tasks promptly and accurately, adapting to the changing needs of the IoT ecosystem. To evaluate the effectiveness of the proposal, we conducted case studies in three IoT application systems: a smart lighting system, a smart water heater, and a room temperature monitoring system. The results demonstrated that the proposed system significantly outperformed traditional methods, providing more timely, context-specific guidance and improving overall maintenance experiences for users.