Rate limits

Limits

Limits apply per account: every key of the account draws from the same buckets. Windows are one minute.

Discovery responses are also cacheable (Cache-Control: public, max-age=300), so cache them rather than calling them per job.

Active jobs

Separately from request rates, an account can have a limited number of unfinished jobs at once (50 by default). Creating more returns 429 TOO_MANY_ACTIVE_JOBS with Retry-After: 30 and a body that includes maxActiveJobs and activeJobs. Each clip of an independent batch counts as one job. If your workload needs more concurrency, contact us.

Headers

Every response to a request that passed key authentication (success or error), and every discovery response, carries:

HTTP/1.1 429 Too Many Requests
Retry-After: 12
X-RateLimit-Limit: 30
X-RateLimit-Remaining: 0
X-RateLimit-Reset: 1760356800

Backing off

On 429 (and 503), wait exactly Retry-After seconds and retry the same request, with the same Idempotency-Key for creates. Without a Retry-After (a 500), use exponential backoff with jitter.

import random, time
import httpx

def request_with_retries(client: httpx.Client, method: str, url: str, **kwargs) -> httpx.Response:
    for attempt in range(5):
        resp = client.request(method, url, **kwargs)
        if resp.status_code in (429, 503) or (resp.status_code == 409 and "Retry-After" in resp.headers):
            time.sleep(float(resp.headers.get("Retry-After", "1")))
            continue
        if resp.status_code >= 500:
            time.sleep(min(8, 0.5 * 2 ** attempt) * (1 - 0.25 * random.random()))
            continue
        return resp
    return resp

The SDKs do this for you, honouring Retry-After up to 60 seconds. THROTTLED (Retry-After: 3600) is not something to wait out in a loop; see Errors.

Staying under the limits

• Poll with GET /jobs/{jobId}?include=none every 5 to 15 seconds, not faster, or use webhooks.
• For many URLs, one POST /jobs/batch with up to 100 clips uses one job-creation request.
• To list work, page through GET /jobs?state=... rather than polling each job.