(eng): We are concerned with design of decentralization control strategy for stochastic systems with global performance measure. Such a strategy is required e.g. for creation of a distributed control systems for traffic light signalization. The technique of Fully Probabilistic Control Design (FPD) is capable of solving the task for centralized control strategy via Kullback-Leibler divergence (KLD) minimization. In this contribution, we propose a way for decentralization of FPD using the Variational Bayes (VB) approximation. This approximation is a well known technique of Bayesian estimation and it is based on approximation of the true posterior density by a product of conditionally independent densities, VB-marginals. The fundamental principle is again minimization of the KLD between a product of conditionally independent densities and the true posterior. Typically, the solution is found in the form of a set of implicit equations.