Abstract Accretion disks of active galactic nuclei (AGNs) have been proposed as promising sites for producing both (stellar-mass) compact object mergers and extreme mass ratio inspirals. Along with disk-assisted migration, ambient gas inevitably accretes onto compact objects. In previous studies, it was commonly assumed that either an Eddington rate or a Bondi rate takes place, although they can differ by several orders of magnitude. As a result, the mass and spin evolution of compact objects within AGN disks are essentially unknown. In this work, we construct a relativistic supercritical inflow–outflow model for black hole (BH) accretion. We show that the radiation efficiency of the supercritical accretion of a stellar-mass BH (sBH) is generally too low to explain the proposed electromagnetic counterpart of GW 190521. Applying this model to sBHs embedded in AGN disks, we find that, although the gas inflow rates at Bondi radii of these sBHs are commonly highly super-Eddington, a large fraction of inflowing gas eventually escapes as outflows so that only a small fraction accretes onto the sBH, resulting in mildly super-Eddington BH absorption in most cases. We also apply this model to neutron stars (NSs) and white dwarfs (WDs) in AGN disks. It turns out to be difficult for WDs to grow to the Chandrasekhar limit via accretion because WDs are spun up more efficiently to reach the shedding limit before the Chandrasekhar limit. For NSs accretion-induced collapse is possible if NS magnetic fields are sufficiently strong to keep the NS slowly rotating during accretion.