9.8 Quotas, Cost Guards, and Provider Limits

A tenant can stay under its per-minute limit all day and still use far more than its plan includes. At acme's 300,000 TPM, a full day of steady traffic is 300,000 x 1,440 minutes = 432 million tokens, while the Pro plan includes 5 million tokens per day. The previous lesson built the per-minute side for acme: a token bucket in a Lua script that reserves input plus max_tokens and refunds the unused part after the provider reports actual usage, and a sorted set that caps the requests in progress.

This lesson covers the limits that work over longer periods and across tenants