Echo state networks (ESNs) are recurrent models whose stability and memory properties are often summarized by the echo state property (ESP). This thesis develops a continuous-time ESN framework in the language of non-autonomous dynamical systems using the skew-product formalism. Building on this setting, we introduce an input-wise notion of multistability through an echo index, which measures the number of attracting solutions induced by a fixed input signal. We provide sufficient conditions guaranteeing echo index one, including contraction- type regimes and large-input regimes, and study robustness under perturbations of the input signal.