The "native" file format is just a zipped collection of CSV tables with LongText, File/Image, Code content saved as zip substreams so it can be browsed simply in the File Manager.
GS-Base can perform intense multicore field value and pivot table calculations so it loads and keeps all the data in RAM. For example, a table with 150 million rows/records with four text fields containing up to several characters each with calculated fields and formulas can use around 15GB of RAM.
Like in classic SQL databases, binary objects can be stored as links to the original disk files, but even if they are stored as zip64 substreams in the database, e.g. hundreds of thousands of "Memo" fields can be loaded almost instantly.
Recent updates: The maximum number of fields is now 16,384 (the max. number of records is still 256 millions). You can apply up to 16,384 field filters at the same time. Users can predefine how individual columns in the loaded text/CSV files are to be treated. New field validations and conversions (when entering the data) based on the regex. New UI commands to perform permanent and dynamic table joining easier. Easy splitting and normalization of data tables (that is, the reverse of the above) with a few clicks. Added support for the SQLite *.db files.