A new approach has been developed to reconstruct the magnetospheric configurations from data by directly representing the magnetic field as a sum of divergence-free contributions from a set of diffuse bubble-like field sources, more or less evenly distributed over the modeling region. Unlike the earlier suggested method based on the radial basis functions, the new fully local approach does not involve toroidal and poloidal fields, which makes it conceptually simpler and computationally more straightforward. The method has been tested on artificial data sets generated using an empirical model and a global magnetohydrodynamic simulation. The new model was shown to faithfully reconstruct the high-latitude magnetospheric structures, such as the field-aligned currents and the diamagnetic depression in the outer polar cusp funnels. The new method can serve as a flexible tool with a tunable resolution and varied focus area, to be used in the data-based modeling of the magnetosphere.