The paper introduces SchNet4AIM, a modified version of the SchNet architecture designed to predict local quantum chemical descriptors, including atomic and interatomic properties. SchNet4AIM is developed to address the challenge of balancing accuracy and interpretability in computational chemistry, where machine-learned models can accurately predict molecular properties but lack physical interpretability. The authors test the performance of SchNet4AIM by predicting a wide range of real-space quantities, such as atomic charges, delocalization indices, and pairwise interaction energies. The model demonstrates high accuracy and speed, breaking the computational bottleneck that has prevented the use of real-space chemical descriptors in complex systems. The group delocalization indices, derived from physically rigorous atomistic predictions, are shown to provide reliable indicators of supramolecular binding events, contributing to the development of Explainable Chemical Artificial Intelligence (XCAI) models. The paper also highlights the computational efficiency and extrapolation capabilities of SchNet4AIM, making it suitable for large-scale chemical simulations and complex chemical processes. The interpretability of SchNet4AIM's predictions is further demonstrated through a case study on the CO2 capture and release by a Calix[4]arene, where the model's predictions on electron delocalization are used to understand the driving forces behind the binding events. Overall, SchNet4AIM provides a robust and physically coherent approach to predicting local quantum chemical properties, offering valuable chemical insights without compromising prediction accuracy.The paper introduces SchNet4AIM, a modified version of the SchNet architecture designed to predict local quantum chemical descriptors, including atomic and interatomic properties. SchNet4AIM is developed to address the challenge of balancing accuracy and interpretability in computational chemistry, where machine-learned models can accurately predict molecular properties but lack physical interpretability. The authors test the performance of SchNet4AIM by predicting a wide range of real-space quantities, such as atomic charges, delocalization indices, and pairwise interaction energies. The model demonstrates high accuracy and speed, breaking the computational bottleneck that has prevented the use of real-space chemical descriptors in complex systems. The group delocalization indices, derived from physically rigorous atomistic predictions, are shown to provide reliable indicators of supramolecular binding events, contributing to the development of Explainable Chemical Artificial Intelligence (XCAI) models. The paper also highlights the computational efficiency and extrapolation capabilities of SchNet4AIM, making it suitable for large-scale chemical simulations and complex chemical processes. The interpretability of SchNet4AIM's predictions is further demonstrated through a case study on the CO2 capture and release by a Calix[4]arene, where the model's predictions on electron delocalization are used to understand the driving forces behind the binding events. Overall, SchNet4AIM provides a robust and physically coherent approach to predicting local quantum chemical properties, offering valuable chemical insights without compromising prediction accuracy.