App Environment Variables (app.yaml)
These variables are defined in app.yaml and injected into the app runtime. Placeholder values (e.g., __GSO_CATALOG__) are patched before deployment — by notebooks/install.py into its generated workspace source folder (recommended path), or in place by deploy.sh (local terminal path).
MLflow Tracing
| Variable | Value | Description |
|---|
MLFLOW_TRACKING_URI | databricks | MLflow tracking server (Databricks workspace) |
MLFLOW_EXPERIMENT_ID | __MLFLOW_EXPERIMENT_ID__ | Experiment for tracing LLM calls. Workspace-specific; validated at startup, cleared if invalid |
LLM Model
| Variable | Value | Description |
|---|
LLM_MODEL | __LLM_MODEL__ | Databricks model serving endpoint for analysis and the create agent. Default: databricks-claude-sonnet-4-6 |
SQL Warehouse
| Variable | Source | Description |
|---|
SQL_WAREHOUSE_ID | valueFrom: sql-warehouse | SQL Warehouse ID, pulled from the app resource named sql-warehouse |
Genie Agent Configuration
| Variable | Value | Description |
|---|
GENIE_TARGET_DIRECTORY | /Shared/ | Where new Genie Agents are created. Override to a specific folder if needed |
Local Development
| Variable | Value | Description |
|---|
DEV_USER_EMAIL | (empty) | User email for local dev auth. Only used when running outside Databricks Apps |
Lakebase PostgreSQL
| Variable | Source | Description |
|---|
LAKEBASE_HOST | valueFrom: postgres | Hostname, injected from the postgres app resource |
LAKEBASE_PORT | 5432 | PostgreSQL port |
LAKEBASE_DATABASE | databricks_postgres | Database name (standard Lakebase default) |
LAKEBASE_INSTANCE_NAME | __LAKEBASE_INSTANCE__ | Lakebase Autoscaling project name (patched by deploy script) |
Auto-Optimize (GSO Engine)
| Variable | Source | Description |
|---|
GSO_CATALOG | __GSO_CATALOG__ | Unity Catalog for optimizer state tables. Patched from .env.deploy |
GSO_SCHEMA | genie_space_optimizer | Schema within the catalog for GSO tables (fixed name) |
GSO_JOB_ID | __GSO_JOB_ID__ | Databricks Job ID for the optimization DAG. Patched from bundle deploy state |
GSO_WAREHOUSE_ID | valueFrom: sql-warehouse | SQL Warehouse for GSO queries |
Deploy Configuration Variables (.env.deploy)
These variables apply to the local terminal path only — they are read by deploy.sh and written by install.sh. They are not injected into the app runtime directly; the deploy scripts use them to patch app.yaml placeholders and configure resources.
On the recommended notebook path there is no .env.deploy file. The equivalent values come from notebooks/install.py widgets (app_name, catalog, warehouse_id, lakebase_mode, lakebase_project_name).
| Variable | Required | Default | Description |
|---|
GENIE_WAREHOUSE_ID | Yes | — | SQL Warehouse ID (hex string from warehouse URL or detail page) |
GENIE_CATALOG | Yes | — | Unity Catalog name (you need CREATE SCHEMA permission) |
GENIE_APP_NAME | No | genie-workbench | Databricks App name (must be unique in your workspace) |
GENIE_DEPLOY_PROFILE | No | DEFAULT | Databricks CLI profile name |
GENIE_LLM_MODEL | No | databricks-claude-sonnet-4-6 | LLM serving endpoint for analysis |
GENIE_LAKEBASE_INSTANCE | No | <app-name> | Lakebase Autoscaling project name (auto-provisioned by deploy) |
How Variables Flow
- The installer collects values — notebook widgets on the notebook path, or
install.sh writing .env.deploy on the terminal path
- The installer patches
__PLACEHOLDER__ strings in app.yaml. The notebook path writes the patched copy into its generated workspace source folder and leaves the Git folder's app.yaml untouched; the terminal path patches the file in place
databricks apps deploy (or the notebook's equivalent SDK call) uploads the patched app.yaml- The Databricks Apps platform injects env vars into the running container
valueFrom variables (e.g., LAKEBASE_HOST, SQL_WAREHOUSE_ID) are resolved from app resources at runtime