Ambient Intelligence (AmI) carries out a futuristic vision of living environments which are sensitive and responsive to the presence of people and, by taking care of their desires, intelligently respond to their actions improving their comfort and well-being. Typically, AmI frameworks are based on distributed context-aware approaches that, by using collections of invisible and interconnected devices, elicit and analyze environmental knowledge for delivering an appropriate set of services. Emotion-aware AmI (AmE) enhances the conventional idea of intelligent environment by exploiting theories from psychology and social sciences for suitably analyzing human emotional status and achieving a higher users' satisfaction. This work proposes a novel approach of combining emotion-aware idea with a neuro-fuzzy framework to train a collection of intelligent FML-based agents aimed at delivering efficient, personalized and interoperable emotion services in an AmE environment. As will be shown in experimental results, where a usability study and a confirmation of expectations test have been performed, the proposed approach is capable of anticipating user's requirements and improving the performance of a conventional AmI framework.