Raynaud and Gruson showed that there is a reasonable algebro-geometric notion of family of discrete (infinite-dimensional) vector spaces. The author introduces a notion of family of Tate spaces ("Tate" means "locally linearly compact") and claims that it is local. The definition takes in account that the K_{-1} of a ring is not necessarily zero. However, we prove that K_{-1} always vanishes after Nisnevich sheafification. As a discrete counterpart of families of Tate spaces, we introduce the notion of almost projective module. We discuss the notions of dimension torsor and determinant gerbe of a family of Tate spaces. The above technique has two different applications. First, we clarify the structure of the ind-scheme of formal loops of a smooth affine manifold Y. This allows to define a "refined" motivic integral of a differential form on Y with no zeros, which is an object of a triangulated category rather than an element of its K_0 group. Second, we show that almost projective modules and families of Tate spaces appear naturally in the study of the cohomology of a family of finite-dimensional vector bundles on a punctured smooth manifold. The canonical central extension that comes from this cohomology allows to interpret the "Uhlenbeck compactification" of the stack of vector bundles on the projective plane as the fine moduli space of a certain type of generalized vector bundles.