This paper describes a method to obtain in a certain sense the best of two well known simulation approaches: Stochastic simulation of chemical reactions and Entity-centered (multi-agents) systems. The stochastic simulation algorithm and its enhanced versions are adapted to simulate the behaviour of well stirred mixture of reactants when spatial localisation is not relevant; Conversely, in entity-centered methods, the causes and consequences due to spatial localisation are well handled. In terms of computational efficiency, the SSA methods are limited by the number of reactions and not by the amount of reactants, whereas it is the opposite for the entity-centered method. We will show how we can benefit from each approach, and also propose a new optimised algorithm for approximate stochastic simulation.