// the take
A SQLite-specific SQL parser, formatter, static analyzer, and language server, built directly from SQLite's own grammar and tokenizer rather than a generic SQL dialect. It's for developers who want editor tooling (LSP, formatting, schema-aware linting) or who need to embed real SQLite parsing into other tools via the Rust/C/Python/WASM bindings.
Parses from SQLite's actual parse.y grammar and C tokenizer instead of reimplementing SQL from scratch, so it handles version gating and compile-time flags (e.g. SQLITE_ENABLE_MATH_FUNCTIONS, RETURNING added in 3.35.0) that generic SQL parsers miss entirely — and it's validated against ~396k statements from SQLite's own upstream test suite at 99.7% agreement, which is a real correctness bar, not a marketing number. The analyzer works schema-aware without a live DB connection and reports multiple diagnostics per pass. The AST retains comments and whitespace, which actually matters for migration/codegen tooling instead of being a throwaway tree. One core (C parser/tokenizer + Rust layer) is exposed consistently across Rust, C, Python, and WASM/JS, plus a dialect mechanism (demoed with Perfetto) for reusing the grammar machinery for other SQLite-derived languages.
Still explicitly 0.x — the README states APIs and CLI surface may change before 1.0, so anything built against libsyntaqlite or the C API should expect breakage. It's effectively a single-maintainer project (830 stars but only 18 forks suggests little outside contribution or independent validation yet). The author discloses heavy AI involvement in the mechanical implementation; that's disclosed honestly, but for a parser whose core correctness claim rests on grammar fidelity, it's worth more scrutiny than the author's 'I understand all of it' framing implies. There's no mention of a security audit for the C tokenizer/parser layer — if you're feeding it untrusted or generated SQL (not just your own editor buffer), the memory-safety story of that C code isn't addressed anywhere in the docs.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →