Echocardiography remains the most widely used cardiac imaging technique, leveraging ultrasound video data to evaluate heart structure and function. While artificial intelligence holds promise for enhancing precision, reproducibility, and efficiency in echocardiographic analysis, existing AI systems are largely limited to single-view, single-task applications. These narrow models fail to integrate complementary information across multiple views captured during a complete echocardiogram study, restricting their clinical utility and performance. To overcome these limitations, researchers have developed EchoPrime, a multi-view, view-informed, video-based vision-language foundation model trained on over 12 million echocardiogram video-report pairs. Leveraging contrastive learning, EchoPrime learns a unified embedding space that captures both common and rare cardiac conditions across all standard echocardiographic views. The model incorporates a view-classification mechanism and a view-informed anatomic attention module to dynamically weight video-specific embeddings, accurately reflecting the anatomical relationships between imaging views and cardiac structures. EchoPrime employs retrieval-augmented interpretation to synthesize information from all videos within a comprehensive echocardiogram study, enabling holistic clinical assessment. This approach allows the model to contextualize findings across views, mimicking the integrative reasoning used by expert cardiologists. Evaluated across five independent healthcare systems worldwide, EchoPrime achieves state-of-the-art performance on 23 diverse benchmarks assessing cardiac anatomy, function, and pathology. It significantly outperforms both specialized task-specific models and earlier foundation models in accuracy, robustness, and generalization. Rigorous clinical validation confirms its ability to support physicians in automated preliminary interpretation of comprehensive echocardiograms, reducing workload and enhancing diagnostic consistency. EchoPrime represents a major advancement in AI-driven cardiac imaging, demonstrating the potential of vision-language models to deliver integrated, clinically meaningful insights from complex multimodal data. Its development marks a shift from isolated, view-specific analysis toward a unified, context-aware approach to echocardiographic interpretation.