Skip to main content
Last updated on

Manage budgets for Unity AI Gateway

Public Preview

This feature is in Public Preview.

Budgets help you monitor monthly Unity AI Gateway spend using filters applied to billing records. You can scope a budget specifically to the Unity AI Gateway product to track only Unity AI Gateway spending, or include Unity AI Gateway in an account-wide budget.

note

Unity AI Gateway budgets also track Genie product usage. To set budgets for Genie specifically, see Manage budgets and cost controls for Genie.

Unity AI Gateway endpoint tags are propagated to the custom_tags column in system.billing.usage, allowing budgets to be scoped to specific teams, projects, or cost centers. For example, if an endpoint is tagged with team=ml-platform, you can create a budget filtered on team=ml-platform to track spend associated with that endpoint’s usage.

For details on creating and monitoring budgets, see Create and monitor budgets.

Requirements​

How budgets work with Unity AI Gateway​

When creating a budget, you can scope it to Unity AI Gateway in two ways:

  • Product scope: Select Unity AI Gateway as the resource type to track only Unity AI Gateway spending (Pay-Per-Token and ai_query inference).
  • Tag and workspace filters: Add tags such as team, project, or cost_center to filter spending to specific endpoints.

You can combine both approaches to create budgets for specific teams or projects within Unity AI Gateway.

Budget configuration with tags

Per-user spend thresholds: For budgets scoped to Unity AI Gateway, you can configure per-user thresholds that track individual user spending and send alerts when a user exceeds their threshold.

Limitations​

  • Budgets with resource tags scope usage to system.billing.usage records that include those tags in custom_tags.
  • Only AI Gateway endpoint tags are propagated to custom_tags in system.billing.usage.
  • Budgets do not support request tags.
  • Unity AI Gateway budgets currently track Pay-Per-Token (PAYGO) and ai_query (batch) inference. Provisioned throughput and external-model inference are not currently tracked.

Next steps​

On this page