The workbench for Looker and Dataform.
A native Mac app that compiles, validates, and runs your BigQuery pipeline, built for you and your coding agent.
Install
Version 0.1.0, signed and notarized. Requires macOS 13 or later.
brew tap graver-app/tap
brew trust graver-app/tap
brew install --cask graver
Or download the DMG — signed and notarized,
no Homebrew needed. It won't update itself; brew upgrade will.
Why this exists
- Doesn't the Dataform cloud IDE already do this?
- Parts of it, well. Auto-compile as you type, run a single object, results right under the code. But it lives in a browser tab, blind to the coding agent doing the writing, and it stops where Looker begins.
- Doesn't Dataform have AI built in?
- It ships with Gemini, which offers itself at every turn and can't see past Dataform's walls. It knows nothing about your Looker project, the half of the pipeline your warehouse exists to serve. Graver's answer is different: bring the agent you already trust, and let it see everything.
- And the Looker IDE?
- Open a second file and the first one closes. Drag-and-drop reports can't be version-controlled, and LookML has almost no tooling around it. It isn't where real work can happen.
- Why one app for both?
- Because it's one pipeline. The warehouse code exists to feed the reports, and the reports break when the warehouse changes. A change is only correct when both sides agree, so the tool should see both sides.
- Where does the coding agent fit?
- It writes most of the code now; the editor is where you read. What the agent lacks is eyes: a way to compile, validate, run, and see what happened. Graver gives the agent the same workbench it gives you.
✓Dry-run everything
Validation that understands pending upstream changes. Schema breaks surface before anything runs, at zero cost, with the query bill estimated up front.
✓Run where you work
Any table, view, or assertion executes natively against BigQuery. Results land in a grid right under the code.
✓Eyes for your agent
Claude Code connects over MCP. Every workbench capability is also a tool the agent can call, and you watch it work.
✓Made for the stack
LookML and SQLX in one window: the whole pipeline your reports depend on, end to end.