Serverless AI inference infrastructure for open-source models with US/EU sovereignty across language, vision, audio, and multimodal