This vision paper presents a new generation of multimodal streaming systems that embed Multimodal Large Language Models (MLLMs) as first-class operators, enabling real-time query processing across multiple modalities. While recent work has integrated …