LogiX-GIN is an innovative Graph Neural Network (GNN) model designed to make graph-based decisions directly interpretable as logic rules. Unlike post-hoc methods, LogiX-GIN embeds explainability at the core of its architecture, ensuring that explanations are faithful to the model’s internal computations. Experimental results show that, unlike other self-explainable approaches which are limited to identifying the most important part of the input, LogiX-GIN provides rules that explain its behavior layer by layer. And despite the constraints imposed on the architecture to ensure this explainability, it achieves performance comparable to black-box models. The introduction of LogiX-GIN opens new perspectives for the use of GNNs in domains where trust and transparency are crucial, such as computational biology, chemistry, and critical decisionmaking systems.