Test a budget before enforcing it
Documentation for Tutela v3.9.0.
Check how a proposed AI call would be handled by a budget before applying the control broadly. A preview evaluates supplied estimates; it does not prove that a later request was executed, billed or blocked.
Prerequisites
- Permission to configure the relevant budget policy.
- An intended scope, such as workspace, user or session, and a spending or token limit agreed with its owner.
- Recent usage and pricing information for the provider and model you plan to test.
- A test scope that does not interrupt other users.
Preview a budget
- Open AI Spend and review the selected period, pricing coverage and missing usage. Resolve unavailable reporting before choosing a limit from those totals.
- In Budgets, review the budget for your intended scope or create the policy through the available budget controls. Record the scope, limit, period and original settings before editing an existing policy.
- Preview a proposed call with its estimated tokens or cost. Record the estimated amount and the action shown by the preview.
- Compare a call that fits the remaining budget with one that would exceed it. Confirm that the returned allow, warn or block action matches the configured policy. If the preview cannot establish a result, resolve its error before enabling the policy.
- In the isolated test scope, send a small permitted request through the intended client. Compare the recorded policy decision and resulting usage with the preview. Do not generate a large bill to test the upper limit; use a small test limit instead.
- Restore any temporary limit when testing is complete. Before applying a new limit to other users, review the result and expected effect with the budget owner.
When preview and usage disagree
A preview uses an estimate before usage is recorded. A later provider response can report a different number of tokens. Missing pricing, delayed usage, a different budget scope or a changed policy can also explain a difference.
Compare the model, scope, period and timestamps first. Inspect the request's Activity evidence and the matching usage record. Do not disable all budgets to investigate one failing scope. If you cannot reconcile the records, use AI usage and cost problems.
Validate
- The configured budget matches the intended scope and period.
- Below-limit and over-limit previews produced understood results.
- A real request from the intended client was checked separately.
- Temporary test settings were restored and rollout ownership recorded.
Next steps
- Understand estimated and billed costs.
- Include budget exceptions in daily checks.