The present work introduces a way of embedding a combustion chemical system in a neural network, in such a way that it can be used, with considerable CPU time and RAM memory savings, in fluid-flow-simulation codes. The system is composed of four neural networks, with three of them simulating the evolution of the reactive species and one providing density and temperature as a function of composition. The performance in terms of accuracy of the networks is assessed by comparison with the results of the direct integration of the thermochemical system for a large number of random samples. Error measurements are reported, and sample evolutions of the chemical system with both methods are compared. It can be summarized that the results of this exercise are satisfactory, and the CPU-time and memory savings encouraging.