Thanks for your idea. I first thought that simply viewing data might not be useful as many IDEs already handle basic data previews. One of core goals was to access stored metadata directly from the file without necessarily scanning the data. Your idea of potentially mixing current implementation (using pyarrow) with DuckDB is very interesting:)

I upvote this suggestion and add pickle and hd5 to the list.