CI checks your code syntax.
Levee checks what your PR quietly breaks — then opens the PR that fixes it.
Before a dbt model PR merges, Levee traces its full downstream impact through DataHub's MCP Server, blocks the merge if something production-critical breaks, generates the backward-compatible migration SQL, and opens a companion Fix PR automatically — then writes the verdict back onto the dataset in DataHub so the next person or agent inherits the context.
Silent Production Downstream Breaks
A dbt model gets a column renamed. The PR looks clean — tests pass, lint is green. It merges. Three days later, the executive churn dashboard shows null, and the production ML feature pipeline ingests wrong values. Nobody connects the cause to the change because three days passed.
Analytics Engineers & Data Platform Leads
Designed specifically for data teams starting to let AI agents or junior engineers propose dbt model schema alterations, column renames, and refactors on production storage.
Advanced: Manual Setup
For manual webhook configurations or standalone CI/CD pipeline actions.
No Code Installation Needed
If your team has a dbt project on GitHub and a running DataHub instance:
- Go to your GitHub repo Settings → Webhooks.
- Payload URL:
https://<your-tunnel-or-deployed-url>/api/webhook - Content type:
application/json - Events: Select Pull requests.
Native CI Pipeline Action
Add .github/workflows/levee.yml to run checks directly inside GitHub Actions:
name: Levee Lineage Guard
on: [pull_request]
jobs:
guard:
steps:
- uses: actions/checkout@v4
- run: python -m agents.levee_agent