OP here. > Could that also be achieved with a virtual table implementation? In general cases, yes, that would solve the issue. > I would have liked to get some reasons WHY GPL should be done and why with an embedded db engine. Where would the OP see any advantage over Relational DB sql? As for schema generation the gain would be minimal. The part where it can really be handy is a traversal query with pattern matching. In order to achieve it now (especially for complex schema), user has to specify a query with tons of JOINs; whereas in graph-oriented queries he/she would only specify a pattern (with some additional conditions regarding 'vertices' and 'edges' and how deep the connection should be). So the data would be stored the same way as it is stored now, the only things that would change are the query compiler stack, query planner and VDBE. Of course, there is a potential performance risk, but since data is stored still in tables, regular indexes may improve the performance to some degree. As for why with an embedded db engine: Currently there are no good server-less (also one file) graph db candidates that can be used on a wide range of devices. Then I heard about SQL/PGQ work and decided to ask here whether this feature would be welcome.