We provide two nonlinear solutions to the model recovery anti-windup (MRAW) design problem, both of them relying on the definition of a set of nested ellipsoids in the state space of the anti-windup dynamics. Each ellipsoidal set arises from a convenient trade-off between size of the ellipsoid and guaranteed exponential convergence rate induced by the corresponding saturated feedback. The first solution is given by a hybrid selection of the MRAW stabilizer, relying on a natural hysteresis switching mechanism. The second solution corresponds to a Lipschitz but non-differentiable scheduled selection, which essentially smoothens out the discontinuous nature of the nested ellipsoids. We discuss the role of our design architecture and establish a number of important properties induced by the proposed controllers. Their effectiveness is comparatively illustrated on a few example studies.