It uses https://github.com/sqlparser-rs/sqlparser-rs as the parser and lexer. The binder, planner, optimizer and executor are in Python. The optimizer stage only works on the logical plan and the rules are heuristic only.
That code could be copied directly from some real-world examples - sqlparser-rs code looks pretty much exactly the same.
I haven't written up a public roadmap yet as I'm still focused on improving the MySQL and PostgreSQL support. While there is technically a SQLite parser in the main tree, it's substantially lower quality than the others. This is due to the fact that it's generated using Bison and not used by any else in production.
SQLite uses a custom parser generator called lemon[0] to parse SQL queries. Sadly that parser is deeply entwined with SQLite itself; it's not trivial to extract a full AST.
My current plan (still a work-in-progress and by no means final) is to use sqlparser-rs[1] via wasmtime. The AST produced by this crate is very high quality and it supports multiple dialects of SQL.
[0] https://www.sqlite.org/lemon.html [1] https://github.com/sqlparser-rs/sqlparser-rs