Traditional smart IoT monitoring systems are overly reliant on central servers, leading to user concerns about privacy leaks and service disruptions, as well as a rigid interactive experience. To address this predicament, this research proposes an innovative decentralized monitoring system, ""Huan-Yu-Chain,"" aiming to create a truly personalized and secure smart butler experience. The core process of this research includes two major innovations. First, a three-tiered modular base station architecture enables local data processing and flexible system collaboration. Second, to achieve humanized interaction, I conducted supervised fine-tuning on a local Large Language Model (LLM). By generating synthetic data with a teacher model and completing the fine-tuning on GCP, the model can accurately understand user intent, invoke tools, and generate responses in a designated style. Furthermore, in intranet mode, all data is stored locally and disconnected from the external network, fundamentally eliminating privacy risks. ""Huan-Yu-Chain"" integrates a decentralized architecture with a customized LLM, demonstrating that highly intelligent services can be achieved without relying on the cloud. This system provides a solution for smart homes that is highly secure, scalable, and humanized.