Abstract Understanding fault geometry and mechanics is key to interpreting the seismotectonic setting of a region. We analyze data from a 1‐year dense seismic array deployment in Southern Italy to infer focal mechanisms and stress field from low‐magnitude earthquakes using machine‐learning techniques. We derive 289 focal mechanisms, including nearly all events with ML>1.3 ${M}_{L} > 1.3$, yielding a data set comparable to that obtained from 6 years of conventional monitoring. The retrieved mechanisms are consistent with the orientation of the main tectonic structures, and the stress field, inverted from the principal fault planes, indicates a dominant normal faulting regime. We illustrate that dense monitoring enables investigation of spatio‐temporal characteristics of the stress field, showing homogeneity at ∼20 km spatial scale and temporal stability of principal stress orientations. Furthermore, seasonal variations in the stress ratio are observed, that could be modulated by hydrological loading.