Skip to main content
Two engineers. One grammar. Fifteen dialects.

We built an SQL
compiler from scratch.

It parses faster than Rust. It passes the official PostgreSQL and BigQuery compliance test suites. It powers column lineage, type inference, query optimization, and a full dbt/Jinja pipeline. And we're just getting started.

Try dfmt — Free SQL FormatterLearn about dquery

The backstory

We're a two-person team from Norway. We love functional programming, data infrastructure, and building tools that make other engineers' lives better.

Three years ago we set out to solve a problem that had been bugging us for years: SQL tooling is terrible. Every tool handles one dialect. Every lineage tool is a black box. Every formatter mangles your Jinja. And everything is slow.

So we built a compiler. A single declarative grammar definition that generates specialized, JIT-optimized compilers for 15 SQL dialects. Each one produces a fully typed, immutable AST with 9,400+ node types. On top of that: scope resolution, column lineage, type inference, query optimization, error recovery, and a complete Jinja evaluator for dbt projects.

The result outperforms hand-written Rust and C implementations on every benchmark we've run. It passes the official conformance test suites of PostgreSQL's libpg_query and Google's ZetaSQL. It compiles 59 public dbt projects — nearly 10,000 models — without errors.

None of this is open source. It's the foundation of everything we're building. But we wanted to give you a way to experience the quality firsthand.

Øyvind
Øyvind
Ingar
Ingar

The SQL infrastructure problem

In 2025, Fivetran acquired SQLGlot. dbt Labs acquired SDF. Then Fivetran and dbt announced a merger. When it closes, one company will control all the major open-source SQL compiler infrastructure.

If your product depends on SQL parsing, lineage, or formatting — your infrastructure is now controlled by a company that competes with you. And SQLGlot has hit its technical ceiling regardless: Python-only, lossy roundtripping, no error recovery, no type inference.

We built the independent alternative. 15 dialects from one grammar. Lossless roundtripping. Full semantic analysis. Faster than Rust. And available to license exclusively.

Free. Forever.

dfmt — your first look at the engine

dfmt is our SQL formatter. It's the first product built on the Datoria compiler, and it's free — no trial, no signup, no strings attached. We want you to feel the difference a real compiler makes.

  • 15 dialects — Snowflake, BigQuery, PostgreSQL, T-SQL, Spark, DuckDB, and 9 more
  • Native Jinja — evaluates templates through a purpose-built engine, not regex
  • Adaptive — respects your formatting choices while normalizing structure
  • Idempotent — format(format(x)) = format(x), always
  • 11,000 files, 0 errors — benchmarked across 58 public dbt projects

15 Dialects. One Grammar.

Every dialect inherits from a shared grammar core. Add a feature once, it works everywhere.

ANSI SQL
BigQuery
Databricks
DB2
DuckDB
MariaDB
Oracle
PostgreSQL
Presto
Redshift
Snowflake
Spark SQL
SQLite
T-SQL
Trino

Need the engine itself?

If you're building SQL-powered products — IDE plugins, migration tools, governance platforms, data catalogs, AI agents — and you need an SQL compiler that actually works across dialects, runs in microseconds, and comes with semantic analysis built in, we should talk.

The compiler — dquery — is not publicly available. Column-level lineage, type inference, query optimization, scope resolution, error recovery, dbt compilation — the full semantic stack. We're opening early access to a small number of companies.

What we're building next

Best-in-class column lineage and semantic analysis. An LSP server for SQL intelligence in your IDE. An agent API for LLM-powered SQL tools. dfmt is the free preview. The full engine is coming.