<p>Abiotic stress poses a significant threat to cucurbit crop productivity under climate change. While numerous transcriptomic datasets have been generated for cucumber, melon, and watermelon under drought, heat, and salt conditions, these data remain fragmented and difficult to interpret across species and stress types. Here, we present AI-PlaNet, a resource platform for multi-omics integration and machine learning–based discovery of stress-responsive genes in cucurbits. We curated and standardized 328 RNA-seq samples from nine independent studies and constructed co-expression networks to identify stress-associated modules. Through multi-species comparative analysis, we identified 134 conserved hub genes across drought, salt, and heat stress. Integrating metabolomic data further revealed transcription–metabolism coordination, particularly during key time points in drought response. Using gradient boosting models trained on promoter features, network metrics, and expression dynamics, we achieved high prediction accuracy (AUC = 0.94) for classifying core stress genes. SHAP analysis enabled interpretable feature attribution, highlighting ABRE motif density and hub connectivity as key predictors. AI-PlaNet includes an interactive visualization panel that links gene expression, co-expression context, and model-based priority scores. This platform provides a reusable and expandable framework for abiotic stress gene mining in cucurbits and offers valuable resources to support functional genomics and stress-resilient breeding strategies.</p> Graphical Abstract <p></p>