This study presents the design, implementation, and evaluation of a multi-agent system (MAS) for intelligent heating control in residential buildings. The system integrates three reactive agents, manual, seasonal and neural, operating within a Home Assistant platform to optimize thermal comfort while minimizing energy consumption. Through the combination of rule-based control, seasonal patterns, and machine learning predictions, the proposed approach demonstrates significant improvements over conventional thermostat systems. Experimental validation using ESP32 microcontrollers, wireless sensors, and Node-RED automation shows promising results for energy efficiency and adaptive control. The neural agent, trained on historical data using a multilayer perceptron (MLP) classifier, achieves optimal performance with ReLU activation functions, demonstrating the potential for intelligent and user-adaptive heating systems in smart home environments.