# Rate limits

Source: https://docs.revoplyai.com/get-started/rate-limits/

> How many requests flow trigger URLs accept, how the limits answer, and how to retry.

The limits below apply to [flow trigger](/flow-triggers/) URLs (`POST /hooks/{token}`).
Limits for the REST API will be published with it.

| Limit                                        | Counted per                | Answer                                        |
| -------------------------------------------- | -------------------------- | --------------------------------------------- |
| 60 requests a minute                         | flow trigger URL           | `429`, `{"status": "rate_limited"}`           |
| 1,000 requests an hour                       | account, all URLs together | `429`, `{"status": "rate_limited"}`           |
| 120 requests a minute                        | client IP address          | `429`, plain text `Too many requests`         |
| Templates automations may send per day       | account and its projects   | `429`, `{"status": "template_limit_reached"}` |
| 3 templates from automations in any 24 hours | customer                   | `429`, `{"status": "customer_limit_reached"}` |

The minute and hour windows are fixed: they reset at the start of each clock minute and
hour. The per-address limit is a sliding one-minute window that exists to slow down
guessing of URLs; a single integration stays well below it.

Every request to a valid URL counts towards the per-URL and per-account limits, including
requests refused afterwards for another reason (a body that is not JSON, a flow that is
switched off) and repeats of an `Idempotency-Key` already used. Requests refused for their
`Content-Type` or their size are not counted.

The two template limits are checked before a contact or conversation is created, and runs
that are queued but have not sent their template yet count towards them.

## When our counters are unavailable [#when-our-counters-are-unavailable]

If we cannot count requests, flow trigger URLs refuse the request with
`503 service_unavailable` rather than accept it unlimited: each accepted request can end in
a paid WhatsApp template.

## Retrying [#retrying]

* Retry `429` and `503` with exponential backoff, starting at about a minute.
* Send the same `Idempotency-Key` on every retry of one event, so a request that was
  accepted but whose answer you did not receive starts nothing new. See
  [Idempotency](/get-started/idempotency/).
* Do not retry other `4xx` answers unchanged: they fail the same way every time. The full
  list is in [Responses and errors](/flow-triggers/responses-and-errors/).