Databricks

Overview

DataStrict governs AI on the lakehouse: models served through Mosaic AI Model Serving, and retrieval over data in Unity Catalog.

Policy binds to Unity Catalog identity, so what a model may see and do follows the same governance as the rest of your Databricks estate.

How it works

Requests to Databricks Model Serving pass through the gateway, which evaluates them against policy before the model answers. When a prompt is assembled from lakehouse data, DataStrict applies the masking and access rules tied to that Unity Catalog identity and purpose, and logs each decision to the Ledger.

Connect

Route your serving endpoint and retrieval calls through the gateway, connect it to your workspace, and map Unity Catalog grants onto DataStrict policy.

# datastrict.yaml
gateway:
  provider: databricks
  workspace: https://my-workspace.cloud.databricks.com
  identity: unity-catalog
  ledger: postgres://ledger.internal:5432/datastrict

What you can enforce

• Mosaic AI serving prompts and responses evaluated against policy in real time.
• Lakehouse data masked per Unity Catalog identity before it enters a prompt.
• Agent and tool actions denied unless a policy allows them for that purpose.
• A queryable, hash-chained audit trail of every decision.