What is an AI asset registry?

Summary

• An AI asset registry is a centralized inventory within Unity Catalog that enables organizations to register, discover, secure, and audit AI assets in one place.
• The registry serves as the foundation for both discovery and runtime enforcement of AI governance, allowing teams to publish assets once and reuse them across the organization.
• Key asset types include models, agents, MCP services, skills, and tools. All are governed under the same framework as data assets.
• The registry enables centralized visibility through permissions, lineage, observability, and cost management, eliminating fragmented governance systems.

What is an AI asset registry?
An AI asset registry is a centralized inventory within Unity Catalog that enables organizations to register, discover, secure, and audit AI assets in one place. It serves as the foundation for both discovery and runtime enforcement of AI governance, allowing teams to publish assets once and reuse them across the organization. Rather than managing AI separately from data, the registry extends the same governance framework and role-based access controls organizations already use for data to all AI components.

What assets belong in a registry

The AI asset registry governs multiple types of AI resources:

Asset Type	Purpose	Access Control
Models	Databricks-hosted and external model providers	EXECUTE to use, MANAGE to configure
Agents	AI agents registered as Unity Catalog securables	CREATE to register new agents
MCP Services	Model Context Protocol servers for tool integrations	EXECUTE to invoke tools
Skills	Reusable AI capabilities and workflows	EXECUTE to use
Tools and Connections	Enterprise and external APIs securely integrated	EXECUTE to invoke, CREATE SERVICE to register

Why a centralized registry matters

Single governance model. Rather than building separate governance systems for AI, organizations extend Unity Catalog privileges to AI assets. EXECUTE permissions allow teams to use models or invoke tools, CREATE SERVICE enables registration of new assets, and MANAGE handles administrative configuration. One governance framework covers every asset type.
Visibility at scale. As organizations scale AI, the number of models, agents, and tools quickly multiplies. Most enterprises still govern these assets separately or not at all, leading to fragmented systems and inconsistent access policies. The registry provides administrators with centralized visibility through permissions, lineage, observability, and cost management from a single console.
Discovery and reuse. Instead of every team independently creating integrations with models and tools, the platform team gets a common inventory and developers get a catalog of trusted building blocks. Assets are discoverable in Catalog Explorer alongside data assets, enabling teams to find and use approved resources without rebuilding.
Cost attribution and control. Understanding AI spend is a critical challenge at scale. The registry enables granular cost attribution down to individual teams, applications, or users. Organizations can set budgets, define hard spend caps, and identify anomalous usage before it becomes a surprise invoice.
Runtime enforcement. Unity Gateway enforces runtime controls across model calls, tool invocations, and agent workflows. Every request flows through the gateway, evaluated against policies defined in the registry before execution and logged afterward. This provides both control and audit capability.

How the registry works with Unity Gateway

Unity Gateway acts as the enforcement layer for the registry. When an application calls a model or invokes a tool, the request flows through the gateway, which evaluates it against the policies and permissions defined in the registry. The gateway logs the request, response, token consumption, and cost. This telemetry then feeds observability dashboards and enables continuous monitoring of AI asset usage.
The registry entries themselves live in Unity Catalog, stored alongside data assets. This means IT administrators can manage AI governance using the same tooling, audit interfaces, and RBAC frameworks they already know.

Use cases

Scaling trusted AI. Teams can register vetted models and tools in the registry, setting policies for who can use them and under what conditions. Developers discover and reuse these approved assets, reducing duplicative integration work.
Compliance and audit. Regulations like SOX, GDPR, and others require audit trails for decision-making systems. The registry maintains a queryable log of who accessed which models, when, and what they were used for.
Cost management. Finance and platform teams monitor AI spend by project, team, or application. If usage spikes unexpectedly, alerts and spend caps help prevent runaway costs.
Multi-tenant governance. In large organizations or managed services, the registry enables platform teams to create and enforce AI policies across multiple tenants with fine-grained access controls.

FAQs

How is an AI asset registry different from a model registry?

A model registry typically tracks model versions and metadata. An AI asset registry is broader, encompassing models, agents, tools, skills, and connections. It integrates asset governance with runtime enforcement through Unity Gateway.

Do I have to register all my models and agents?

No. You can continue to use unregistered models and tools, but they won't benefit from centralized governance, observability, or cost tracking. As your organization scales AI, registering critical assets helps maintain control.

Can I use the registry with external model providers?

Yes. External models accessed through API keys or OAuth can be registered as model services in the registry. The gateway handles authentication and access control.

How does the registry handle different access levels?

The registry uses Unity Catalog privileges. EXECUTE allows users to call a model; READ lets analysts view lineage and logs. MANAGE is reserved for administrators configuring policies and resource limits.

What happens when a registered asset reaches a spend cap?

When an asset or team reaches its spend cap, further requests are denied until the cap is reset or adjusted. The system logs the denial and can trigger alerts to notify the team.