A slope digital twin is a virtual slope model that is able to continuously, even in real-time, learn from actual observations (e.g., monitoring data, slope performance records, and site investigation data) obtained from its physical counterpart to enhance the performance of the slope model. This study proposes a practical framework to develop a slope digital twin and describes its application to predict the temporal variation of rainfall-induced slope instability of a real slope in Hong Kong. When compared with a conventional slope model that remains unchanged, the proposed slope digital twin combines monitoring data (e.g., data on rainfall and pore water pressure in the slope) and slope survival records to probabilistically update the model. Specifically, the most suitable model settings are selected, and both the hydraulic and strength parameters of the soils are updated, thereby decreasing the associated uncertainties. The updated slope model can predict pore water pressure responses of a target rainfall consistent with the actual measurements. Furthermore, the model can be used to predict the temporal variation of slope stability (e.g., by using a factor of safety with quantified uncertainty or slope failure probability) during the target rainfall. Because the monitoring data and past slope survival records are incorporated in the model updating, the proposed slope digital twin enhances the prediction of soil hydraulic responses and slope stability. The predicted temporal variation of slope stability agrees well with the observed slope failure induced by an extreme rainstorm in June of 2008.