Skip to main content

​
Availability

Governed Metadata is available from v2.2.0. In the Community Edition, the metadata pages are not in the admin sidebar.

​
Overview

Governed Metadata is a set of standard fields that you add to LLMs, tools, and data sources. Administrators decide which fields each object must have. AI Studio checks the values when you save an object. Examples of fields:
  • Business, technical, and application owners
  • Lifecycle state and expiration date
  • Risk tier and data classification
  • Regulatory applicability and approved consumers
  • Support contact
Administrators define the fields once. AI Studio then validates the values on each object, records each change, and reports the objects that are not compliant. Governed Metadata applies to these objects:
  • LLMs
  • Tools
  • Data sources
  • Plugin resource types that opt in. Refer to Plugins.

​
Vocabularies

A vocabulary is a controlled list of terms. Each term has a value, a label, an optional description, and a Deprecated flag. A field of type vocabulary accepts only the terms of its vocabulary. A deprecated term is still valid, but it causes a warning. To manage vocabularies, go to Governance > Metadata vocabularies. You cannot delete a vocabulary that a schema uses. The API returns 409 in this case.

​
Schemas

A schema is a named set of fields. It applies to one or more object types, or to all object types. To manage schemas, go to Governance > Metadata schemas. Edit Metadata Schema form with an enforced schema for LLMs Each field has these settings: Use Required to publish to separate the person who submits an object from the person who releases it. The submitter saves a draft without the field. The reviewer adds the value when they activate the object.

​
Enforcement

Each schema has an enforcement level:
  • Advisory: AI Studio records validation problems on the object and shows them in the compliance report. It does not block a save.
  • Enforce: AI Studio checks each create and edit of an LLM, a tool, or a data source. The save fails with 422 until all required fields with the severity Error are valid. This applies to the admin console and the admin API.
More than one active schema can apply to the same object type. AI Studio merges them. The required fields of an advisory schema stay warnings in the merged schema. Only the fields of an enforced schema block a save.
Start with an advisory schema. Change it to Enforce only after the compliance report shows no problems. If you enforce a schema first, edits to existing objects fail until someone adds the missing values.

​
The Governance Core Schema

On the first start, the Enterprise Edition creates an advisory schema named Governance Core. It applies to all object types. It has these fields: AI Studio also creates the four vocabularies that these fields use: The regulatory_applicability field uses the regulatory_framework vocabulary. Each other vocabulary field uses the vocabulary with the same name. You can edit the schema and the vocabularies, or add more schemas.

​
Enter Metadata on an Object

When a schema applies, the LLM, tool, and data source forms show a Governance Metadata section. AI Studio validates the values while you type. When a save fails, the form marks the fields that are not valid. The detail page of the object shows the stored values and the validation status. Governance Metadata section of the LLM form after a failed save

​
Compliance Report

Go to Governance > Metadata coverage. AI Studio validates each object again against the current schemas. The report puts each object in one of these groups:
  • Missing
  • Invalid
  • Expired: a date with Warn when in the past is in the past
  • Warnings
  • Valid
Each row has a link to the object, so you can fix it. The report also lists the instances of plugin resource types that opt in, and MCP servers. Metadata coverage report with valid, invalid, and missing objects

​
Where the Values Go

A change to a field with Sent to gateways changes the configuration snapshot. The Edge Gateways in the namespace then show Pending until you push the configuration. A change to any other field does not need a push.

​
Change History

AI Studio records each change to the metadata of an object, and who made it. A change from a plugin shows as plugin:<id>.

​
Permissions

Governed Metadata uses the metadata resource of role-based access control: To read or set the metadata of an object, you need the permission for that object, not for metadata. For example, a user who can edit LLMs (llms:write) can also set the metadata of an LLM. The default Editor role does not have metadata:publish.

​
Plugins

Plugins can use Governed Metadata in these ways:
  • Read and write metadata with the management service API. Refer to Governed Metadata in the Service API.
  • Check or change values before AI Studio saves them with object hooks on the governed_metadata object type. Refer to Object Hooks.
  • Add schemas and vocabularies from the plugin manifest. AI Studio creates them as inactive and advisory. An administrator activates them. Refer to Governed Metadata Contributions.
  • Opt a resource type in with supports_metadata: true. The instances of the type then use the object type plugin_resource:<plugin_id>:<slug>. A plugin can write plugin_resource:self:<slug> for its own types.
  • Show the metadata form in a plugin UI with the <governed-metadata-fields> and <governed-metadata-badges> web components. Refer to Governed Metadata Web Components.
  • Read the gateway-visible fields of an LLM in a gateway plugin. Refer to Governed Metadata in the Plugin Context.
A plugin that publishes its own objects, such as prompts or agents, keeps them in its own storage. AI Studio does not validate these objects for the plugin. For plugin objects, enforcement works only if the plugin calls SetObjectMetadata on save and refuses the save when the call returns ok=false.