DuckDB Now Ships inside dbt v2
dbt v2, powered by a new Rust engine, ships with DuckDB support built-in instead of requiring a separate Python package. This lets data teams run dbt locally without external databases or warehouses.
- DuckDB adapter is now bundled with dbt v2 and automatically downloads its driver on first run, eliminating the need to install dbt-duckdb separately
- v2 adds catalog support for DuckLake and Iceberg, native SQL comprehension with column-level lineage, and stores metadata as queryable Parquet files instead of massive JSON manifests
- The Rust-based engine compiles to a single binary, making local development faster and enabling the VS Code extension to catch errors in the editor before queries hit the warehouse
dbt v2, released September 14, 2026, now ships with DuckDB built in as a native adapter instead of requiring a separate Python package. The Rust-based Fusion engine that powers v2 automatically downloads and caches the DuckDB driver on first run, eliminating setup friction. This matters because dbt-duckdb has been popular since 2021 for letting data teams run SQL transformation projects locally without paying for cloud warehouses—the GitHub issue requesting DuckDB support in Fusion got 146 reactions, signaling real demand from the community.
v2 adds capabilities the old Python adapter lacked. You can now define DuckLake and Iceberg catalogs in a `catalogs.yml` file with catalog-aware materializations, letting dbt handle the ATTACH statements automatically. The metadata layer changed too: instead of writing massive JSON manifest files, v2 can export project metadata as Parquet files to `target/info_schema/v1/`, which you can query directly with DuckDB. This is practical for audits and CI checks—you can ask questions like which models are tables versus views or which models lack tests without loading hundreds of megabytes of JSON or spinning up the warehouse.
The Rust binary also enables faster local development and smarter error catching. The v2 engine understands SQL natively across multiple dialects, so it catches invalid column references and type mismatches before queries hit the database. The same analysis produces column-level lineage locally without needing a dbt platform account. v2 pins a specific DuckDB version rather than relying on pip's resolver, which enabled read-write Iceberg support and let dbt move some adapter logic into native DuckDB extension functions. The VS Code extension gets the same SQL comprehension for autocomplete and inline errors as you edit.