This paper presents a novel, multi-scale, framework, for the simultaneous analysis of multiple data streams, called StreamStory. The framework models the data streams as a hierarchical Markovian model by automatically learning states and transitios, and aggregating them into a hierarchy of Markov chains. This approach aims to compensate the gap between lowlevel streaming observations and high-level output/alerts which provide a value for higher levels of streaming data analysis, like inference and prediction, and provides ground for qualitative interpretation of the data.