What is an API Rate Limit?

Limits are usually expressed as a number of requests per second, minute, or day, and providers commonly return a `429 Too Many Requests` response along with headers indicating the current limit, remaining quota, and when it resets, once a client exceeds its allowance. Well-behaved clients read these headers and back off proactively rather than hammering the API and hoping. Rate limits exist for good reason: without them, a single misbehaving client (or a genuine traffic spike) could degrade service for every other consumer sharing the same backend. From the client side, handling rate limits gracefully means implementing exponential backoff and retry logic, and — for systems making many calls on behalf of many users — often means queuing or throttling outbound requests locally so the shared limit is never hit in the first place. For AI agents that can generate a burst of tool calls very quickly (far faster than a human clicking through a UI), rate-limit awareness is not optional — an ungoverned agent loop can exhaust a provider's quota in seconds, breaking the integration for every other tenant sharing that same API credential.