I'm currently working on this, building a new incremental, multi-file ready source connector to solve some of these limitations with the current file-source. However, the design I have at the moment still retains a one table per connection, primarily because: • we want to allow (optionally) a schema to be provided rather than relying on inference. • we want to allow glob-style pattern matching to determine which files belong to a table. • we want to allow granular (table-level) settings for what to do in cases of missing or new columns, or malformed rows. Providing these could get a bit unwieldy in one connection for multiple tables but would be keen to understand more about your use-case @Arnis Dizdari :)