// the find
SQLMesh/sqlmesh
Scalable and efficient data transformation framework - backwards compatible with dbt.
SQLMesh is a data transformation framework that competes with dbt, adding proper environment isolation, incremental model tracking, and a plan/apply workflow borrowed from Terraform. It targets data engineering teams that have hit dbt's walls around efficiency and environment management. Now under the Linux Foundation.
Virtual Data Environments let you create dev branches without duplicating tables — it uses views over shared physical data, which actually solves the dbt cost problem in a principled way. The plan/apply model gives you a real diff of what will change before you run it, including column-level lineage impact. SQL dialect transpilation via sqlglot means you can write in one dialect and execute in another, which is useful when migrating warehouses. Unit test generation from live query results is a genuinely useful shortcut — no hand-writing YAML fixtures.
3,133 stars is thin for infrastructure you're betting production pipelines on — the ecosystem of community plugins, adapters, and StackOverflow answers is tiny compared to dbt's. The commercial Tobiko Cloud product is not clearly separated from the open-source layer in the docs, so it's hard to know what you're actually getting for free versus what requires a sales call. State management (snapshots, intervals, environment metadata) is stored in your data warehouse, which means a messy failed migration can leave you with corrupted state and no clean rollback story. Python model support exists but feels secondary — the docs lean heavily SQL-first, and the Python execution model has sharp edges around environments and dependencies.