Credits and limits

A video that returns a result costs 10 credits for every started minute of it, however many rows come back. A 26-minute video is 260 credits; a 26-minute-and-one-second video is 270. Each response reports what it charged, so you never have to infer it.

A job that succeeds but returns zero rows is charged. The extraction ran; the narration simply did not contain what the schema asked for.

The batch ceiling is enforced: more than 200 URLs in one POST /v1 is refused with batch_too_large. The per-minute limit is enforced too, across every key in a workspace over a sliding one-minute window. Every response carries X-RateLimit-Limit and X-RateLimit-Remaining; past the limit you get 429 rate_limited with a Retry-After header — honour it rather than looping, see retries and idempotency. One batch call carries up to 200 videos, so the limit is really about polling. A webhook removes the need to poll at all, and every paid plan includes them.

A finished result is returned to you exactly once. The first successful read hands over the rows and deletes them and the submitted URL from our side in the same request. Read the job again and you get status: done with a null result, which is not an error — you already have the data.

Everything else about the job — its status, timestamps and character and chunk counts — is kept, which is what the usage and billing screens are built from.