Integrations — Transformation
NerveStax and dbt
Agents read your dbt project, change models, and open a pull request you review.
What we do
What NerveStax does with dbt
NerveStax reads your dbt project and its artifacts — manifest, run results and catalog — into one record of models, tests, sources and runs, with model-level lineage and freshness. Agents write and change models, run dbt against a development target inside a sandbox, and open a pull request your team reviews and merges.
-
Reads your project and its runs
The manifest, run results and catalog become one record of models, tests, sources and runs. Every conversation starts from what is already known about the project, instead of re-reading it.
-
Model-level lineage and freshness
What a model depends on, what sits downstream of it, when each asset last built, and where the project has drifted from the tables actually in the warehouse.
-
Writes and changes models
An agent edits SQL and YAML in a checkout of your repository, adds or fixes tests, and lists the downstream models and schedules the change touches before you read it.
-
Runs dbt on a development target
dbt run, test and build execute inside the sandbox using the project's configured connection. Mark a project protected and a person approves before any dbt command runs against it.
-
Checks models against your schedules
One tag on a model says how often it should build. Agents read that schedule, flag tags that resolve to nothing, and explain why a model ran when it did. They never change a schedule.
-
Opens a pull request you approve
Work lands on its own branch and arrives as a pull request in GitHub or GitLab with results and reasoning. Agents cannot merge, so your review rules decide what ships.
Background
About dbt
dbt is the transformation tool data teams use to build warehouse models as SQL and YAML in a git repository, with tests, documentation and a manifest describing the project. NerveStax reads a dbt project as it already is.
How it connects
What connecting it looks like
-
01
Connect your repository
Point a project at the repository holding your dbt project and the base branch agents should open pull requests against.
-
02
Add the warehouse connection
Give the project the credentials its dbt profile needs. They are stored encrypted and no screen ever shows them back.
-
03
Let it read the project
NerveStax parses the manifest and, after a run, the run results and catalog, building the record of models, tests, lineage and freshness.
-
04
Mark production projects protected
A protected project requires a person's approval before any dbt command runs against it. Unprotected projects run on their configured target.
Scope
What this does not cover
- Lineage is model-level. Column-level lineage is not built.
- dbt Core projects, read as they are. There is nothing to restructure and no dbt Cloud account involved.
- dbt commands use the connection configured on the project. Protection, not a separate identity, is what keeps an agent off a production target.
Questions
dbt and NerveStax
Q01Does NerveStax work with dbt Core?+
Yes. NerveStax reads a dbt Core project from your git repository as it stands — models, tests, sources, the manifest and the artifacts a run produces. It does not need a particular folder layout, a dbt Cloud account or a migration. The dbt version your project already pins is what runs in the sandbox.
Q02Do I need dbt Cloud?+
No. NerveStax runs dbt itself, inside a sandbox, against the connection on your project. Your orchestration stays where it is — usually Airflow. If you do run dbt Cloud or another platform alongside, nothing about that stops NerveStax reading the project and opening pull requests.
Q03Does it run dbt against production?+
Not without a person. dbt commands run in the sandbox against the target the project configures, which is normally a development target. Mark a project protected and every dbt run, test and build on it stops for approval first. Changes still reach production only through a pull request someone merges.
Q04How deep does the lineage go?+
No. Lineage today is model-level: what a model reads from, what reads from it, and what a change would touch downstream. Column-level lineage is not built, and we would rather say so than imply it. Freshness, run history and drift against the warehouse are there.
Q05Where does the agent run dbt?+
In a sandbox created for that conversation, with a checkout of your repository — not on your laptop and not on a CI runner. The branch is the durable record; the sandbox is a cache that is destroyed when the conversation goes idle.