What is Unity Gateway and how does it relate to Unity Catalog?

Summary

• Unity Catalog is the governance layer; Unity Gateway is the API gateway. Unity Catalog manages access control, discovery, lineage, and auditing for all data and AI assets in your organization. Unity Gateway is the platform for securely routing requests to AI models and external APIs.
• Unity Gateway applies Unity Catalog policies. When you route AI model calls and external API requests through Unity Gateway, those requests are governed by the access controls, audit logging, and policy enforcement defined in Unity Catalog.
• They are complementary layers. Unity Catalog covers data governance (who can access which tables, columns, and assets). Unity Gateway covers AI/API governance (who can call which models and services, with what parameters, at what rate).
• Together they protect both data and AI. Unity Catalog ensures only authorized users and agents can read sensitive data. Unity Gateway ensures only authorized applications can invoke models or call external services, logging every interaction for compliance and troubleshooting.
• AI agents need both. Agents that query data must be governed by Unity Catalog's policies on that data. Agents that call external tools or models must route through Unity Gateway, where their requests are logged and controlled.

What is Unity Gateway and how does it relate to Unity Catalog?

Unity Gateway and Unity Catalog are distinct but complementary layers of Databricks security and governance:

• Unity Catalog is the unified governance platform for all data and AI assets. It manages access control, metadata, discovery, lineage tracking, and audit logging for tables, files, models, agents, dashboards, and other assets. It defines who can access what and records every access.
• Unity Gateway is the AI API gateway. It securely routes requests to foundation models, external APIs, and custom AI services. It handles authentication, rate limiting, request validation, and logging for all AI model calls.

The relationship: Unity Gateway applies Unity Catalog governance policies. When a user or agent sends a request to an AI model through Unity Gateway, Unity Catalog's access controls determine whether that request is allowed, and both systems log the interaction for compliance and auditing.

Why they work together

A modern data and AI stack needs both layers:

Layer	What It Controls	Examples
Unity Catalog	Who can access data and AI assets	A data scientist can read the customer table; an analyst cannot access sensitive columns; an agent can query approved sources
Unity Gateway	Who can invoke AI services and external APIs	A business application can call the LLM endpoint; a user cannot exceed rate limits; all requests are logged

Without Unity Catalog, data access is uncontrolled and unaudited. Without Unity Gateway, AI requests lack governance. Together they create an enterprise-grade security model.

How they integrate

Unity Catalog defines identity and permissions. An agent running in Databricks has a service principal identity (assigned via Databricks) and permissions (managed by Unity Catalog). When that agent:

1. Queries a table: The request is checked against Unity Catalog's access policies. If allowed, the data is returned; if denied, the query fails and the attempt is logged.
2. Calls an external model through Unity Gateway: The request is authenticated using the agent's identity, checked for rate limits, and routed to the model. Unity Catalog's audit log records the invocation.

Both systems share the same identity and permission model, so governance is consistent across data access and AI invocations.

Key differences from the user's perspective

Aspect	Unity Catalog	Unity Gateway
What it protects	Data tables, files, models, dashboards, metadata	AI model endpoints, external API calls
Main concern	Data privacy, lineage, discovery, compliance	API security, rate limiting, request validation
Where you configure it	Workspace settings, object permissions, tags, policies	Gateway settings, route definitions, rate-limit rules, API keys
What gets audited	Data access (who read/wrote which table/column)	AI requests (who called which model, when, how many times)

Use cases that need both

• Agents querying databases: An agent retrieves customer data (governed by Unity Catalog) and passes it to an LLM (routed through Unity Gateway). Both steps are audited.
• Shared models across teams: A foundation model endpoint is registered in Unity Catalog and routed through Unity Gateway, so access is controlled and usage is tracked.
• Compliance auditing: Audit logs from both systems show the complete flow: user → agent → data access (UC) → model call (UG) → response.

Getting started

• For data governance: Set up Unity Catalog in your workspace to manage access to tables, files, and assets. Use the three-level namespace (catalog.schema.object) to organize resources.
• For AI governance: Configure Unity Gateway to route requests to AI models and external APIs. Define who can call which endpoints and set rate limits.
• For agents: Enable agent service principals in Databricks, assign them permissions in Unity Catalog, and route their model calls through Unity Gateway.

FAQs

Do I need both to use Databricks?

Yes, for enterprise security. Agents and applications that interact with both data and AI need governance at both layers. If you only access data, Unity Catalog alone suffices. If you only call models externally, Unity Gateway alone suffices. For comprehensive AI workflows, both are required.

Can I use Unity Catalog without Unity Gateway?

Yes. Unity Catalog governs all data and AI assets. If you don't route AI requests through an external gateway, you still get data governance, discovery, and lineage from Unity Catalog alone.

What if I call an AI model directly without Unity Gateway?

Direct calls bypass Unity Gateway's rate limiting and validation but are still audited if the caller's identity and permissions are managed by Databricks and Unity Catalog. For enterprise control, use Unity Gateway.

How are audit logs from both systems connected?

Both log the user or agent identity, timestamp, and action. The shared identity model ensures you can trace a request from data access through AI invocation in audit logs.