How do I set AI budgets by team, user, application, or agent?

Summary

• Budgets in the Unity Gateway track estimated AI spend against thresholds you define and are generally available; create them in account settings under Usage → Budgets.
• Scope by team, user, application, or agent using resource and Unity Catalog tags — only endpoints carrying matching tags count toward a budget.
• Shared and per-user thresholds. Set a pooled monthly limit plus a per-user monthly limit, with per-user overrides for specific users or groups.
• Alert or block. When a threshold is reached, send an email alert (access continues) or block further Unity Gateway usage until the budget resets or an admin raises the limit.
• Full cost attribution. Every request is logged to Unity Catalog system tables by identity, workspace, endpoint, tags, model, and provider, with a cost analytics dashboard.

How do I set AI budgets by team, user, application, or agent?

Databricks lets you put spend controls on AI usage through budgets in the Unity Gateway. A budget tracks estimated AI spend against thresholds you define and can scope to a team, user, application, or agent using tags — so you can attribute and cap cost at the level of granularity your organization needs. Budgets are generally available and are created in your account settings.

Why Databricks AI Gateway budgets help control AI spend

• Create a budget in a few steps. In account settings, go to Usage → Budgets, create a budget, and select Unity Gateway as the resource type. Optionally scope it to specific workspaces.
• Scope by tags. Apply resource and Unity Catalog tags — for example, a team, application, or agent tag — so only endpoints carrying matching tags count toward the budget. This is the basis for attributing spend by team, user, application, or agent.
• Shared and per-user thresholds. Configure a shared monthly threshold across everything in scope and a per-user monthly threshold, with per-user overrides that grant specific users or groups a higher limit.
• Choose the action at the limit. For each threshold, send an email alert while access continues, or block further Unity Gateway usage until the budget resets or an admin increases the threshold.
• Pair with rate limits. rate limits cap throughput per endpoint or per user, while budgets cap dollars — use them together as layered controls.
• Attribute every request. Usage is logged to Unity Catalog system tables by identity, workspace, endpoint, tags, model, and provider, and a cost analytics dashboard breaks spend down for chargeback and forecasting.

Getting started

• Read Get started with Unity Gateway budgets.
• Create a budget under Usage → Budgets and select Unity Gateway as the resource type.
• Tag endpoints by team, application, or agent, then set shared and per-user thresholds with alert or block actions. See the AI Gateway budgets reference.
• Add rate limits and review the cost analytics dashboard.

FAQs

What can an AI Gateway budget be scoped to?

A budget can be scoped by workspace and by resource and Unity Catalog tags — such as a team, application, or agent tag — so only endpoints with matching tags count toward it.

What happens when a budget threshold is reached?

Depending on how you configure the threshold, Databricks either sends an email alert while access continues, or blocks further Unity Gateway usage until the budget resets or an admin raises the limit.

Are budgets an exact spending cap?

No. Budgets use near-real-time cost estimates, so actual spend can vary around the configured threshold; they are best used together with rate limits as layered controls.