Abstract-Pervasive computing calls for developing distributed infrastructures featuring large-scale distribution, opennes, context-awareness, self-organisation and self-adaptation. There, it is quite natural to see services (software functionality, data, knowledge, signals) as spatial concepts: they are naturally diffused in the network, and in each location they are sensitive to the context and compete with each other-as such, they can be active in one or multiple regions (niches) of the network.To support and engineer this scenario, we propose a natureinspired coordination model of chemical-inspired tuple spaces. They extend standard tuple spaces with the ability of evolving the "weight" of a tuple just as it represented the concentration of a chemical substance in a biochemical system, namely, in terms of reaction and diffusion rules that adaptively apply to tuples modulo semantic match. We show that this model can be used to enact self-* properties in pervasive systems, through typical spatial patterns involving computational fields, paths, and segregation.