Unity AI Gateway: Runtime Governance for All Your Enterprise AI

Unity AI Gateway: Runtime Governance for All Your Enterprise AI

PUBLIC PREVIEW · DAIS 2026

Every AI model call in your company costs money. Do you know how much? Who made it, for what purpose, and whether it followed your security policies?

Unity AI Gateway is Databricks’ runtime governance layer: a single endpoint that controls AI spend, enforces policies, and delivers full observability across all your models, agents, and MCP servers — regardless of provider.

The problem driving uncontrolled spend

AI adoption is accelerating faster than governance can keep up. With that acceleration come three frictions organizations didn’t expect:

Unmanaged spend

AI agents are now the fastest-growing source of enterprise data spend — and most organizations have no visibility into how much they spend per model, team, or application.

Ungoverned behavior

Organizations struggle to control what agents do, what data they expose, and how they behave in production against real users.

Fragmented ecosystems

Traditional governance approaches were not built for multi-agent environments or ecosystems based on MCP servers and third-party frameworks.

Unity AI Gateway was built to solve exactly this: a runtime governance layer that sits between your apps and all your AI providers, enforcing policies centrally regardless of the model or agent being called.

What is Unity AI Gateway?

Unity AI Gateway is a runtime governance layer for enterprise AI. It helps organizations govern AI spend, AI behavior, and AI adoption across models, agents, MCP servers, AI tools, and AI skills — all from a single control point that works wherever your AI runs.

Apps & Agents

Unity AI Gateway

Authentication · Policies · PII Redaction · Rate Limits · Cost Caps · Logs

FMAPI
OpenAI
Anthropic
Bedrock
Azure OAI

How it works, step by step

1
An app or agent calls the single Unity AI Gateway endpoint

All calls go through one entry point regardless of the target model or provider.

2
Unity Catalog authenticates the caller

Unity Catalog identifies who is making the call (user, service, agent) and verifies their permissions before proceeding.

3
The policy engine acts at runtime

Applies PII redaction, content filters, per-principal rate limits, and hard cost caps before routing the request.

4
Routing to the selected model with automatic failover

The request is routed to the configured model (FMAPI, OpenAI, Anthropic, Bedrock, Azure OpenAI) with automatic failover if the primary endpoint is unavailable.

5
Every call logged to UC system tables

Complete record for audit, per-team or per-app chargeback, and analysis with tools like Lakewatch.

Three core capabilities

Total spend control

AI spend governance with hard budgets, rate limits, spend attribution, and cost visibility across providers, teams, and applications.

Runtime governance

Service policies, guardrails, approvals, and observability across models, agents, and MCP servers — applied before the call ever reaches the model.

Open ecosystem

Govern OpenAI, Anthropic, Gemini, open-source models, agents, MCP servers, and AI frameworks across providers from one single pane of glass.

Unity AI Gateway vs. Unity Catalog: what’s the difference?

Unity Catalog

Governs AI assets: registered models, datasets, features, lineage, and data access permissions.

Defines the policies and stores the context that Unity AI Gateway enforces at runtime.

Unity AI Gateway

Governs AI interactions at runtime: every call to a model, agent, or MCP server made by an app or agent.

Enforces in real time the policies defined in Unity Catalog before the call reaches the model.

Supported providers

Unity AI Gateway supports all major model providers natively, with automatic failover between them:

FMAPI
Databricks
OpenAI
GPT-4o / o3
Anthropic
Claude
Bedrock
AWS
Azure OpenAI
Microsoft

You can also govern open-source models, custom agents, third-party MCP servers, and any AI framework (LangGraph, CrewAI, Agno) through the same endpoint.

Frequently asked questions

References

  • #Databricks
  • #Unity AI Gateway
  • #AI Governance
  • #Unity Catalog
  • #MCP
  • #Agents
Share:
Contáctanos