Mirror symmetry is one of the most mysterious dualities in mathematics. Roughly, it predicts that given any Calabi-Yau variety, there exists a mirror Calabi-Yau variety such that a rich list of geometric relations hold between the two, involving Hodge numbers, Gromov-Witten invariants, variation of Hodge structures, Floer homology (Fukaya category), coherent sheaves, stability conditions and so on. Despite continual progress in the subject, a fundamental question remains unclear: to what extent do mirrors exist, and how to construct the mirror variety? Here we propose a new approach to answer this question, based on latest developments from non-archimedean geometry, in particular the theory of Berkovich spaces, as well as derived non-archimedean geometry. Our goal is to conceive and pursue a full-fledged theory of non-archimedean mirror symmetry, which will lead to new results unattainable from existing methods. We propose to work out a general mirror construction, starting directly from a non-archimedean Strominger-Yau-Zaslow torus fibration, conjectured by Kontsevich-Soibelman, by counting non-archimedean analytic disks with boundaries on SYZ torus fibers. First we need to establish the existence of such counts in full generality, based on non-archimedean Gromov-Witten theory and tail conditions. Then we have to prove various properties of the mirror algebra, including associativity, radius of convergence and singularity estimates. Finally we propose to use wall-crossing formulas to glue local mirror algebras together to obtain the global mirror variety. A long-term goal is to show that the mirror construction is an involution, the best exhibition of mirror duality. We also aim for applications outside mirror symmetry, in particular towards the moduli of KSBA stable pairs in birational geometry. Our project is intimately related to the ongoing Gross-Siebert program based on logarithmic geometry. We also expect fruitful future interactions with their program.