NLP researcher here.
From a simple test, this tool is at the hobby level. The text is not interpreted semantically, because the so-called summarization is nothing but blocks of text from the original text, copied ad-literam. It lacks any kind of rewording. My guess is that for content retrieval it employs the the most rudimentary model one can think of [1], does a plain token indexation, then chooses the blocks with the highest density of non-ordinary words.