Billing and limits

What is counted, how a page is quoted, and the ceilings that apply.

How a page is quoted

Processing is quoted in pages before a compile starts, with the maximum charge shown alongside the estimate. A page count read from the document itself is labelled verified; a count the document only declares — the number Word saved — is labelled declared. A file whose format states no count at all is quoted at nothing: it is named in the preflight with the reason and left out of the total, because a page count derived from file size is an invented number.

Spreadsheets

A spreadsheet is billed on the page count of the sanitized PDF it is converted to, counted after that conversion. Before it, there is no page count: preflight names xlsx, ods and csv files and shows no number for them rather than quoting one from file size. What is charged is settled against the pages the read actually produced, and never above the maximum shown before the run.

Included pages, and what expires

Included pages belong to the billing month they are granted in: when the next month's pages are granted, whatever is left of the previous month expires. Unused pages do not roll over and are not refunded if you cancel. A balance can only be spent while a plan is active. Pages past the included allowance are billed at the published rate of $0.04 per standard page.

Refunds

Refunds: ask within 14 days of payment and the payment is refunded in full, provided fewer than 10% of the plan's included pages have been consumed — 50 pages on Developer, 250 on Team. Past that, the payment is not refunded. Unused pages are not refunded on cancellation. Subject to the terms as updated.

The ceilings

Note TAVONEL processes sources up to 5 MB and 80 pages. A source above either is refused rather than accepted and dropped: the byte ceiling at the capability call with 413 SOURCE_EXCEEDS_PROCESSING_CEILING, the page ceiling after the document is decoded, because intake never decodes it.

How often you may call the API

Every scoped request consumes one unit of a per-minute allowance held per key and per scope, so a key reading documents and asking questions does not spend one budget on both. The window is a clock minute rather than a rolling one: an allowance that is spent is free again at the top of the next minute.

Note Over the allowance the answer is 429 API_RATE_LIMITED. There is no Retry-After header on that code and no X-RateLimit- headers anywhere in the API today — waiting for the next clock minute is sufficient by construction, and publishing a header we do not send would be worse than publishing the window. If the allowance itself cannot be read the request is refused with 503 API_RATE_LIMIT_UNAVAILABLE rather than run unbounded; that is fail-closed behaviour, not a limit you hit. The hourly activation allowance in the table above is separate, answers ACTIVATION_RATE_LIMITED, and does carry Retry-After.