Platform
Rate limits
Each API key may make 600 requests a minute across every /v1 endpoint. Every response reports the key's remaining budget.
The budget is per key, not per address, so one server calling for many projects is not throttled collectively. Every response carries the key's budget:
| Header | Meaning |
|---|---|
X-RateLimit-Limit | Requests per minute this key may make. |
X-RateLimit-Remaining | Requests left in the current window. |
X-RateLimit-Reset | Seconds until the window ends. |
Past the budget a request is 429 RATE_LIMITED with Retry-After (seconds) and X-RateLimit-Remaining: 0. The same headers suffixed -apiKey are still sent for now but are deprecated: read the unsuffixed ones.
Failed authentication
- Requests with no key, a malformed key or a key that never existed are limited to 30 a minute per client address. Past that, the address gets
429 RATE_LIMITEDinstead of401until the block ends. A valid key used from the same address keeps working. - A key that exists but is dead — revoked, expired, its project not active — has its own budget of 30 refusals a minute, so one stale key retrying in a loop never blocks your other keys.
Other limits
Some actions carry their own limits, documented with them: webhook URL changes (20 an hour per project) and test events (5 a minute per endpoint), messages per channel, and calls in progress per project and key. Each answers 429 or its own code with a Retry-After where waiting helps.
Backing off
On 429 and 503, back off exponentially with jitter and never retry sooner than Retry-After. Retry a 5xx write with the same Idempotency-Key until an answer carries Idempotent-Replayed: true, which is final for that key (Idempotency). The SDKs do all of this for you and expose the budget as rateLimit / rate_limit on every response.