Databricks¶

Databricks support is optional and outside the core dependency direction.

python -m pip install "agentmuru[databricks]>=0.3,<0.4"

The executable readiness scenario is examples.databricks_agent:

python examples/databricks_agent.py

It checks package and environment readiness without attempting the network.

Environment¶

Configure the values used by the integration path:

DATABRICKS_HOST
DATABRICKS_WAREHOUSE_ID
DATABRICKS_TOKEN or an approved Databricks SDK credential source
DATABRICKS_VOLUME_URI when the application uses a Unity Catalog volume
DATABRICKS_APP_PORT and DATABRICKS_APP_NAME inside Databricks Apps

Identity handling¶

workspace_client() binds a client to the current operation identity. A user principal requires a forwarded Databricks access token and creates an operation-scoped client. User-scoped clients are never cached across identities.

Application identities can use SDK credential discovery. SQL connections for an application identity may be reused behind a lock after a health check. User SQL connections are created and closed per operation.

Available helpers¶

Workspace service helpers list warehouses, build a serializable catalog tree, and trigger a job. SQL helpers query records, execute statements, and manage transactions. Unity Catalog helpers list catalogs, schemas, tables, schemas, and bounded table rows.

Pass SQL values as parameters. Catalog, schema, and table helpers quote identifiers, but dynamic identity and authorization decisions still belong to the application.

Qualification boundary¶

Offline tests validate identity isolation, input validation, serialization, SQL parameter handling, and missing-configuration failures with fake clients. A deployment must separately qualify credential-backed calls against its workspace, warehouse, Unity Catalog permissions, network policy, and audit requirements.