It runs on the tables/databases defined in Databricks and uses an optimized reader (Photon).

We are also expecting a serverless offering, without the need to create a cluster (so pay-per-use ).

It is not hard to set up so you can test it against a classic cluster.