Very interesting! I played around with it, and it seems to perform abstractive summarization of fairly good quality with coherent output. I can see myself using this. Can I ask what kind of model are you using, and what dataset was out trained on?