Skip to main content
Genie WorkbenchDatabricks App

Genie Workbench

Create, score, and optimize Databricks Genie Agents. A unified developer tool deployed as a Databricks App — with on-behalf-of auth, Lakebase persistence, and a benchmark-driven optimization pipeline.

Get StartedDeployView on GitHub →
12
IQ quality checks
3
maturity tiers
4
optimization pipeline tasks
6
optimization levers

The continuous-improvement loop

Five capabilities, one workflow

Enter the loop at any point and repeat as the Genie Agent evolves — after optimizing, re-scan to see the updated score.

1

Create

A multi-turn AI agent walks you from business requirements to a fully configured Genie Agent.

2

Score

The rule-based IQ Scanner grades Genie Agent quality across 12 checks and assigns a maturity tier.

3

Optimize

A benchmark-driven pipeline measures real accuracy, diagnoses failures, and iterates to a target.

4

Track

Every scan, optimization run, and config change is persisted to Lakebase so you can see progress.

5

Watch

GenieWatch reports per-Agent cost, usage, feedback, and executed-resource lineage from system tables.

Quick start

Install from a notebook

Genie Workbench runs entirely on the Databricks Apps platform. The recommended path is the notebook installer — clone the repo into a Databricks Git folder, set five widgets, and Run All. No local terminal, CLI profile, Node, or uv setup required.

Read the Deployment Guide
notebooks/install.py
# 1 · Clone into a Databricks Git folder
Workspace → Create → Git folder

# 2 · Open notebooks/install.py, set widgets
app_name · catalog · warehouse_id
lakebase_mode · lakebase_project_name

# 3 · Attach Serverless (env v5) → Run All

Start here

Choose your path

Genie Agent developer

Build and refine Genie Agents. Start with the Create Agent and the IQ Scanner.

Workspace admin

Deploy and operate the Workbench. Start with the Deployment Guide and the auth model.

Contributor

Extend the codebase. Start with the Architecture Overview, then the relevant feature doc.