A member holding two keys, each under its own key limit, can still be stopped by their member limit. A squad whose members are each under their individual limits can still be stopped by the squad limit.
Periods
Every limit carries a window it is measured over:
New limits default to monthly. A limit of 0 means no limit at all, which is also the default for a scope you have never configured.
Windows are calendar-based in UTC, not rolling. A daily limit resets at midnight UTC regardless of
when the spend happened, so a team spread across time zones shares one reset moment.
What happens at the limit
Requests from a scope that has reached its limit are rejected with an HTTP 429, and the error names which limit tripped — the key’s own limit, or the member/squad cap above it. The agent sees a rate-limit error and stops; nothing is served at partial quality. Spend through your own BYOK provider keys counts toward these limits as well. A limit caps what a scope spends, not what Edgee bills for. If usage data is briefly unavailable, requests are allowed through rather than blocked. Limits fail open.Watching a limit before it trips
Set a budget alert on any scope that has a limit, and you get a notification at 50%, 80%, or 100% of it. Member and squad budget alerts are percentages of these limits, so a scope with no limit set has nothing for the alert to measure and the rule stays silent. Members can see their own position — spend so far and the limit in force for them — without needing admin access.Limits vs routing budgets
Routing strategies also carry a budget, and it does a different job:
They compose. A common setup is a routing strategy that steps down to cheaper models as spend accumulates, with a usage limit behind it as the backstop. If you want the strategy itself to stop rather than step down, its Block at 100% setting is the strategy-side equivalent of a limit.