Privacy is a very important issue nowadays. While many works deal with data privacy, in this paper, we focus on preserving the privacy of the infrastructure resources on which Ambient Intelligence (AmI) applications are deployed. We argue for an Internet of Things (IoT) based on intelligent agent: IoT- Agent. Our approach allows to preserve the privacy of data related to the resources according to their owners’ desiderata. Our solution is a middleware for the configuration, deploy- ment and monitoring of distributed (AmI) applications in a smart environment. This middleware takes into consideration the heterogeneity and the dynamicity of such environment. This paper presents our solution in terms of: (1) the graphs we use to describe the available infrastructure resources and the deployable application requirements, and graph homomorphisms to model the deployment of applications; (2) a goal-directed Multi-Agent System (MAS) which we show to be a well-adapted paradigm that provides a clear separation between the applicative and the infrastructure layers. The result is an increase in resource privacy and the facilitation of the deployment of AmI applications.