How does an AI gateway integrate with data governance?

Summary

• The Unity Gateway is not a separate governance system — it extends Unity Catalog, the same layer that governs your data, to AI assets like models, tools, and agents.
• AI assets are first-class securables: models, tools, and agents are governed with the same role-based privileges you use for tables — for example EXECUTE to use a model or tool and MANAGE to administer it.
• Every request takes one governed path: each call is authenticated, authorized against Unity Catalog, and routed through the gateway before reaching any external system. See how AI governance works.
• Usage, guardrails, and cost controls are built in: the gateway tracks token usage and latency, logs request/response payloads to governed tables for audit, applies guardrails such as PII detection and content filtering, and enforces per-principal rate limits and budgets.
• The same discovery, lineage, and audit that govern data apply to AI — so policies stay consistent across model and tool providers without duplicate setup.

How does an AI gateway integrate with data governance?

On the Databricks Data Intelligence Platform, an AI gateway integrates with data governance by extending the same governance layer used for data — Unity Catalog — to AI assets. Rather than standing up a separate control system for models and agents, the Unity Gateway treats models, tools, and agents as governed securables and routes every AI request through one authorized, logged path. The result is that the access control, lineage, and audit you already apply to tables now apply to AI.

Why Unity Gateway for governed AI access

• Built on Unity Catalog. The gateway extends Unity Catalog's governance model to AI assets, registering models, tools, and agents as first-class securables. Access uses standard role-based privileges — for example EXECUTE to use a model or tool, privileges to register new ones, and MANAGE to administer configuration and limits — so AI assets are governed with the same permission model as data. See how AI governance works on Databricks.
• One governed path for every request. When an agent or application calls a model or tool, Databricks authenticates the request, authorizes it against Unity Catalog, and routes it through the gateway before it reaches any external system. This gives you a single control plane across Databricks-hosted foundation models, external models, external tools and Model Context Protocol (MCP) servers, and agents.
• Usage tracking and inference logging. Requests, token usage, and latency flow through the gateway and are recorded to governed tables, and request/response payloads can be logged to governed inference tables for monitoring, debugging, and audit. This enables cost attribution down to individual teams, applications, or users.
• Built-in guardrails. The gateway applies guardrails to model interactions, including PII detection, prompt-injection mitigation, and unsafe-content filtering. Service policies (in Beta) can allow, deny, or require approval for individual requests and responses, and can redact or block detected PII. See govern the content of model service interactions.
• Rate limiting and cost controls. The gateway enforces granular rate limits at the principal level, measured in queries per minute and tokens per minute, and supports budgets that monitor and cap spend across model and agent usage. Read new governance capabilities to scale AI agents with confidence.
• Consistent policy across providers. The gateway provides unified API access across models with built-in fallbacks, rate limits, and guardrails, so policies stay consistent without duplicate setup per provider. Read the AI gateway as the governance layer for agentic AI.
• Discovery, lineage, and audit for AI. The same discovery, lineage, and audit capabilities that govern data now extend to AI assets, giving you end-to-end visibility across models, teams, and workflows.

Getting started

• Read how AI governance works on Databricks and the Unity Gateway overview.
• Configure guardrails with govern the content of model service interactions.
• See Unity Gateway is generally available and what's new in Unity Catalog.

FAQs

How does the AI gateway relate to Unity Catalog?

The gateway extends Unity Catalog to AI. Models, tools, and agents are registered as Unity Catalog securables and governed with the same role-based privileges, lineage, and audit as data, so there is no separate governance system to maintain.

What does the AI gateway govern?

It governs access to Databricks-hosted foundation models, external models, external tools and MCP servers, and agents — through one control plane with consistent access control, guardrails, rate limits, and logging.

How does the gateway track usage and cost?

Every request flows through the gateway, which records token usage and latency and can log request/response payloads to governed tables. This supports auditing and cost attribution to specific teams, applications, or users.