> ## Documentation Index
> Fetch the complete documentation index at: https://scrinly.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Host rate limits

> Design reliable clients around host politeness, queueing, and retry behavior.

Scrinly limits traffic per destination host as well as by browser capacity. By default, a host receives no more than two concurrent requests with at least one second between starts. Crawls and individual renders targeting the same host share this budget.

This protects origin servers and makes throughput intentionally different for 100 URLs on one host versus 100 URLs across many hosts. `/render/smart-batch` groups work by domain so the queue can schedule mixed-host workloads efficiently.

## Client retries

For `429` and transient `5xx` responses:

1. Honor `Retry-After` when present.
2. Use exponential backoff with random jitter.
3. Cap attempts and total elapsed time.
4. Avoid retrying validation, authentication, or ownership failures.
5. Keep concurrency below the level that produced throttling.

Do not create a duplicate async job merely because status polling was delayed. Continue polling the original ID unless its submission definitively failed.