OCR API for historical documents

Send one page image, get back its text exactly as written. The same models that read the documents in ArchiveLM, without the app around them. Built for the pages general-purpose OCR gets wrong: old type, handwriting, and languages other than English.

Pricing

You're billed for successful pages only. A page we fail to read returns 502 ocr_failed and isn't billed, though it does count toward your monthly page limit.

What it's good at, and what it isn't

Quickstart

1. Sign in and open Dashboard → OCR API. Create a key. It's shown once, so store it. To go past 100 pages a month, add a card on the same page.

2. Send a page.

curl -X POST https://www.archivelm.com/api/v1/ocr \
  -H "X-API-Key: alm_..." \
  -H "Content-Type: application/json" \
  -d '{"image_url": "https://example.org/scans/page-001.jpg", "language_hint": "German"}'

Response:

{
  "request_id": "…",
  "text": "Berliner Abendpost …",
  "chars": 8412,
  "pages": 1,
  "model_tier": "standard",   // "enhanced" when a second, stronger pass was needed
  "retried": false
}

Endpoints

Base URL: https://www.archivelm.com/api/v1. Authenticate with the X-API-Key header.

Limits

• JPEG, PNG, WebP or a single-page PDF. Type is checked from the file, not the name.
• image_url: HTTPS, up to 10 MB, no redirects. Use it for anything over ~3 MB.
• image_base64: up to ~3 MB of image (the encoded request must stay under 4.5 MB).
• Two requests at a time per account. A call can take a few minutes when the second pass runs, so set your client timeout to 300 s.
• Errors: 400 bad input, 401 bad key, 402 free pages used up (add billing), 413 file too large, 415 not JSON, 429 too many concurrent requests or monthly cap reached, 502 page not read (not billed), 503 busy or daily capacity reached.

Your documents

We keep nothing. The image is read, the text goes back to you, and both are discarded. We log only metadata: when you called, file size, how long it took and whether it worked. Pages are read by Google's Gemini models (see our subprocessors).