I self-host Collabora (online version of LibreOffice; OnlyOffice is another option), and my data lives on my NAS, but it could just as easily live on S3 or some distributed data store.

Oh this is interesting. Any pitfalls you could talk about before I go popping this up myself?

Business that stole everyone’s information to train a model complains that businesses can steal information to train models.

Yeah I’ll pour one out for folks who promised to open-source their model and then backed out the moment the money appeared… Wankers.