OCR (document text)

POST /v1/ocr returns the text of a PDF or an image, page by page, in Markdown. A page that already has a text layer (a PDF exported from a word processor) is read directly; a scanned page or a photo is read by optical character recognition.

POST https://api.realmoon.ai/v1/ocr

• Authentication: an sk-moon-… key with the inference scope (see Authentication).
• Body: multipart/form-data (with the Content-Length header) or JSON, in the shape of the Mistral OCR API.
• Response: the text of each page read.

The only model is moon-ocr-1. It runs on our servers in France: the document is not sent to any provider and is not kept once the response is returned.

Parameters#Link to the Parameters section

Multipart#Link to the Multipart section

JSON#Link to the JSON section

In JSON, any other parameter is rejected with 400 unknown_parameter.

Pages#Link to the Pages section

• Without pages, the whole document is read, up to 50 pages. Beyond that, the API responds with 400 too_many_pages: choose the pages, and split a long document across several requests.
• An image has a single page, page 0.
• A page requested more than once is read only once.

Example#Link to the Example section

The OpenAI SDKs have no method for this endpoint: call it over HTTP.

curl https://api.realmoon.ai/v1/ocr \
  -H "Authorization: Bearer $MOONAI_API_KEY" \
  -F model=moon-ocr-1 \
  -F file=@invoice.pdf \
  -F pages=0,1

import os
import requests

with open("invoice.pdf", "rb") as document:
    response = requests.post(
        "https://api.realmoon.ai/v1/ocr",
        headers={"Authorization": f"Bearer {os.environ['MOONAI_API_KEY']}"},
        data={"model": "moon-ocr-1", "pages": "0,1"},
        files={"file": ("invoice.pdf", document, "application/pdf")},
        timeout=100,
    )
response.raise_for_status()
for page in response.json()["pages"]:
    print(page["index"], page["source"])
    print(page["markdown"])

import fs from 'node:fs';

const form = new FormData();
form.append('model', 'moon-ocr-1');
form.append('pages', '0,1');
form.append('file', new Blob([fs.readFileSync('invoice.pdf')], { type: 'application/pdf' }), 'invoice.pdf');

const response = await fetch('https://api.realmoon.ai/v1/ocr', {
  method: 'POST',
  headers: { Authorization: `Bearer ${process.env.MOONAI_API_KEY}` },
  body: form,
});
const result = await response.json();
for (const page of result.pages) console.log(page.index, page.source, page.markdown);

The same call in JSON, with the document in base64:

curl https://api.realmoon.ai/v1/ocr \
  -H "Authorization: Bearer $MOONAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"model\": \"moon-ocr-1\", \"document\": {\"type\": \"document_url\", \"document_url\": \"data:application/pdf;base64,$(base64 -w0 invoice.pdf)\"}, \"pages\": [0, 1]}"

Response#Link to the Response section

{
  "model": "moon-ocr-1",
  "pages": [
    {
      "index": 0,
      "markdown": "# Invoice No. 2026-118\n\nDate: November 4, 2026\n\n| Item | Amount |\n| --- | --- |\n| Services | €1,200.00 |",
      "source": "text",
      "images": [],
      "dimensions": null
    },
    {
      "index": 1,
      "markdown": "Payment terms: 30 days.",
      "source": "ocr",
      "images": [],
      "dimensions": null
    }
  ],
  "usage_info": { "pages_processed": 2, "doc_size_bytes": 184320 },
  "usage": { "pages": 2, "pages_not_read": 0, "credits": 0 }
}

The credits value is made up.

Costs#Link to the Costs section

The price is fixed per page read; it appears in GET /v1/models?type=ocr (pricing.credits_per_unit, billing: flat) and in the “Models” tab of the portal. The request holds the price of the requested pages, then charges only for the pages actually read. An unreadable document or a request that fails costs nothing.

Limits#Link to the Limits section

The reading service is shared: when other requests keep it busy, the API responds with 429 too_many_concurrent or 503 ocr_busy. For a large document, send a few pages at a time and run the requests one after another.

Errors#Link to the Errors section

Key, balance and spend cap errors are in Errors.

See also#Link to the See also section

• Privacy: what Moon AI keeps, and does not keep, from a document.
• Models: the max_pages and max_file_mb fields.