# Rate limits

Limits keep the service fast for everyone and keep us within the limits of the registries and sites we
check on your behalf. A refused call is never charged.

## Per key

| What | Limit |
| --- | --- |
| Requests with one API key | 10 a second, with bursts up to 50 |
| Calls with `fresh=true` | 30 a minute per key, with bursts up to 10 |
| Bulk and crawl jobs that use a browser | 10 an hour per account |
| Free tools | 60 a minute |

Going over answers 429 `rate_limited` with a `Retry-After` header in seconds. Wait that long and try
again.

## Registries

WHOIS and availability checks go to the registry that runs each TLD, and many registries allow only a few
queries a minute. We share each registry's allowance fairly between all callers and never go over it. When
a registry is busy, a lookup waits up to two seconds; if it still can't go, it answers 502
`source_unavailable` with a `Retry-After` header and costs nothing. Jobs wait longer and use half the
allowance, so single lookups stay quick.
[TLD coverage](/docs/tld-coverage) says which registries are tighter than usual.

## Busy sites and browsers

Website checks fetch each site at most 120 times a minute across all our users, so a popular site can't be
overloaded through us. When a site is at that limit, the check answers 503 `site_busy` and asks you to
retry in 30 seconds. Checks that run a real browser share a small pool; when it is full you get 503
`browser_busy` and can retry a minute later. Neither is charged.

## Free checks without an account

The website lets visitors without an account run 5 free checks a day. That allowance
belongs to the website and isn't available through the API: every API call needs a key.

## Handling limits well

- Read `Retry-After` and wait at least that long before retrying.
- Retry with the same `X-Request-Id`, so a call that did go through is never charged twice.
- For many targets, send one [bulk job](/docs/bulk-and-jobs) instead of many single calls: it paces
  itself within every limit.
