Budgets and limits

Control budgets and request rates.

Configure budgets

Allocate Organization → Team → Person → API Key budgets in USD. Spending plus unused child allocations must fit within the parent. Blank shares the parent remainder; zero blocks spending. Budgets are cumulative, without monthly resets. Each person belongs to at most one team. After leaving, personal budgets remain independent. Unrestricted business uses a separate unlimited organization and team.

Request limits

Set RPM and TPM on organization, team, user and API Key pages. Fixed child allocations sum within each parent; blank intermediates pass fixed descendants upward. Blank shares unreserved capacity and zero blocks calls. UTC calendar minutes reset counters; admission atomically counts the unique path and 429 consumes no capacity. TPM uses the incoming estimate rather than final output usage. Multiple instances need shared Redis.

Handle rejections

Inspect key and related entity settings, then logs. Changes require management permissions. Displayed balances are not provider invoices or guarantees about concurrent settlement.

Budget enforcement

Check all applicable cumulative budgets along Organization → Team → Person → API Key; they do not reset monthly. Blank shares the parent’s available balance and zero blocks spending. A person may join only one team; after leaving, the personal budget remains and subsequent usage no longer counts toward the old team. Remaining budget is not a reservation: concurrent calls, streams and long tasks may exceed the limit before usage settles.

Cost controls

Respect Retry-After on 429, use bounded exponential backoff and avoid nested retry layers. Reduce history and output, separate application keys and inspect snapshots. Deleting a key does not refund historical spend.