SQLBuild, dbt, and SQLMesh are all SQL pipeline frameworks. They share common ground but differ in design philosophy and feature focus.Documentation Index
Fetch the complete documentation index at: https://docs.sqlbuild.com/llms.txt
Use this file to discover all available pages before exploring further.
Feature comparison
| Feature | SQLBuild | dbt | SQLMesh |
|---|---|---|---|
| Testing | |||
| Unit tests with model chaining | Chain across multiple models | YAML-stub, single model | CTE-based, single model |
| E2E scenario tests | Fixture worlds with real graph execution | No | No |
| Local E2E replay | Capture from warehouse, replay in DuckDB | No | No |
| Macro / UDF / table function tests | TEST(mode: macro/udf/table_fn) | No | No |
| Zero-row assertions | __assert__ CTEs in tests and scenarios | No | No |
| Audits | |||
| Built-in audits | not_null, unique, accepted_values, relationships | not_null, unique, accepted_values, relationships | Extensive (statistical, string pattern, etc.) |
| Blocking audits | Block promotion from staging table | Tests run after materialization | Block during plan; during run, data already written |
| Delta/interval-scoped audits | Per-microbatch audit cycle before DML | No | Per-interval for time-range models |
| Compilation | |||
| SQL validation | SQLGlot-based, offline | dbt Fusion (proprietary license) | SQLGlot-based |
| Column-level lineage | Compile-time, fast and rich modes | Post-hoc via docs | Compile-time |
| Column contract validation | Infer columns, check types offline | YAML schema contracts at runtime | Schema contracts via plan |
| SQL transpilation | For local E2E replay into DuckDB | No | For cross-dialect model execution |
| Python macros | @macro() syntax | No (Jinja only) | SQLMesh macro syntax |
| Jinja support | No (Python macros instead) | Yes (core templating) | Yes |
| Incremental | |||
| Incremental strategies | append, delete_insert, merge | append, delete+insert, merge | time-range, unique-key, partition, SCD Type 2, append |
| Microbatch execution | Configurable batch sizes with per-batch audits | Microbatch (recent addition) | Batch size support |
| Stateful interval tracking | No - cursor-based (no external state) | No | Yes - tracks which intervals ran |
| SCD Type 2 models | Coming soon | Snapshots (limited) | Built-in |
| Environments | |||
| Virtual environments | No | No | Pointer swaps, no compute cost |
| Environment diffs | Full row-level data comparison | No | Table diff |
| Zero-copy cloning | sqb clone | No | No |
| Models | |||
| SQL models | MODEL() header with inline config | Jinja-templated SQL + YAML sidecar | MODEL DDL |
| Python models | Coming soon | Limited (remote only) | Pandas, PySpark, Snowpark, BigFrames |
| Custom materializations | Python with full framework hooks | Jinja-based | Python-based custom model kinds |
| dbt | |||
| dbt interop | Run alongside dbt - reads manifest, no migration | N/A | Reimplements Jinja for compatibility |
| Other | |||
| Reference syntax | __ref() - parses as valid SQL | {{ ref() }} - Jinja template | model_name with dependency tracking |
| Adapters | DuckDB, Snowflake, BigQuery, Databricks | 30+ (community adapters) | DuckDB, Snowflake, BigQuery, Databricks, Spark, Redshift, Postgres, Trino, MySQL |
| State requirements | Stateless by default | manifest.json + target/ | Requires state store (local DB, S3, or Valkey) |
| Playground | sqb playground | Clone example repo | Example project |
| AI agent skills | sqb skills update | No | No |
Design philosophy
SQLBuild is test-first. Audits block promotion from a staging table. Unit tests run before their target models. Scenarios exercise real project graphs with fixture data. Local replay runs the same tests in DuckDB without a warehouse connection. The framework is stateless by default - no external database or state store needed to get started. dbt is Jinja-based SQL templating. Unit tests are YAML-defined stubs for single models. Audits run after materialization. Acquired by Fivetran. SQLMesh is state-first. Virtual environments and interval tracking require a state store (local database or PostgreSQL for production). Acquired by Fivetran.Not yet in SQLBuild
- SCD Type 2 models - slowly changing dimension support (in active development)
- Python models - Pandas, PySpark, Snowpark, BigFrames (in active development)
- First-party partition state tracking - opt-in stateful partition tracking (currently possible via custom materializations, first-party support coming soon)
- Broader adapter support - PostgreSQL, ClickHouse, and Microsoft SQL Server are coming soon