Software User Assistance (UA) is afforded from captured User Interface (UI) screen images, with reference to persisted Machine Learning (ML) models. The captured screen images are processed—e.g., using rasterization, Optical Character Recognition (OCR), and/or establishment of a coordinate system—with individual UI elements being determined therefrom. Referencing the persisted ML models, the software application/application state for the captured image is identified. UA data relevant to that application/application state is generated from the model, and then provided to the user (e.g., in a text box overlying the UI screen). Through the capture and processing of UI screen images, embodiments afford a homogenous UA experience for installation, maintenance, and/or upgrade of heterogeneous members of a larger overall landscape, over software lifecycles. Embodiments may be deployed locally on a frontend computer, in order to avoid exporting UI images due to privacy and/or security concerns.