Hi François S., Ontop translates each SPARQL query into one SQL query. When you have only one source, and this source speaks SQL, then Ontop can directly speak to it (as long as it knows its dialect). Otherwise, if you have multiple sources or if your source don't speak SQL (e.g. MongoDB, data lake files or a WebAPI), you need to use a database federator/data virtualization platform that makes all the data looking relational and virtually appearing as coming from "one single source". Denodo is particularly good at that (especially when you have WebAPIs) and there is also open-source alternatives like Dremio and Teiid.
Unfortunately, to the best of my knowledge, these database federators do no support graph databases. So if you need to federate with a SPARQL endpoint, you currently need to do it at the SPARQL level.
