Co-founder here. The 63-char limit still applies (we didn't recompile Postgres!) but we have some code in front, embedded in a layer of PgBouncers, that intercepts the query, parses it and rewrites it into a shorter dataset ID hash that we then "mount" on the database on the fly using Postgres FDWs before forwarding it.

We also use this to drop unwanted queries and rewrite clients' introspection queries (e.g. information_schema.tables) to give them a list of featured datasets instead of normal Postgres schema names.