Nowadays, the overconsumption of artificial sweeteners and their related adverse health impacts have proposed an urgent need to develop safe and healthy alternatives. Herein, we introduce ChemSweet, an AI-based platform for the rapid discovery of potential sweet molecules (http://chemsweet.ddai.tech) with the consideration of their physicochemical properties, sweetness profile, and health risks at the same time. Machine learning prediction models of four important physicochemical and four toxicity properties were established and integrated with the platform to evaluate the candidate molecules' biosafety and stability during the processing processes. Then, a new sweet taste prediction system was developed which ensures the sweet evaluation of six specific kinds of sweeteners. To facilitate the practical application of ChemSweet, the SuperNatural database was integrated for the rational screening of promising new sweeteners. We successfully identified 294 potential sweeteners that simultaneously meet the multiple anticipated criteria. We believe that ChemSweet will serve as a useful tool for identifying safe and healthy sweeteners while reducing the timeframe and high experimental costs.