The authors propose a control architecture aimed at optimizing energy resources for residential prosumers in a demand response framework. The control of resources is performed in two stages. The global control functions, based on the solution of a predictive optimal control problem are ensured by a cloud-based computation platform whereas a closed-loop local controller is responsible for the management of field components and actuators. The methodology is developed for managing residential micro/nano grids comprising PV generation, battery storage and interruptible loads. Simulations are aimed at assessing how much optimality can be affected by prevision errors. Tests are carried out considering a simulated environment characterized by realistic and highly resolved demand and generation curves.