OCR Process

Extract text and structure from documents with arva-ocr in markdown, JSON, or text.

POST https://api.sagea.space/v1/ocr/process

Authentication

Send Authorization: Bearer $SAGEA_API_KEY on every request. Get a key

Parameters

Send a multipart/form-data body with the following fields.

Parameter	Type	Required	Description
`model`	`string`	No	OCR model to use. Default is `arva-ocr`.
`document`	`file`	Yes	PDF, PNG, JPG, TIFF, or WEBP file up to 100MB.
`output_format`	`string`	No	One of `markdown`, `json`, or `text`. Default is `markdown`.
`bbox`	`boolean`	No	Set to `true` to include bounding boxes per block. Default is `false`.
`language_hint`	`string`	No	Hint such as `ne-NP`, `hi-IN`, `en-US`, or `es-ES`.
`pages`	`string`	No	Page range to process, for example first 5 pages. Default is all pages.

Example request

Extract an NPR 24,500 invoice issued to Aarav Sharma.

curl -X POST https://api.sagea.space/v1/ocr/process \
  -H "Authorization: Bearer $SAGEA_API_KEY" \
  -F "model=arva-ocr" \
  -F "document=@invoice_npr24500.pdf" \
  -F "output_format=json" \
  -F "bbox=true" \
  -F "language_hint=en-US"

Works for Hindi and Nepali documents such as citizenship cards and VAT bills.

Response

A successful call returns text, page count, confidence, and layout blocks.

{
  "text": "INVOICE #INV-2042\nBilled to: Aarav Sharma\nTotal: NPR 24,500\n",
  "pages": 1,
  "confidence": 0.98,
  "model_used": "arva-ocr",
  "blocks": [
    {
      "type": "header",
      "page": 1,
      "bbox": [40, 30, 560, 90],
      "confidence": 0.99,
      "text": "INVOICE #INV-2042"
    },
    {
      "type": "table_row",
      "page": 1,
      "bbox": [40, 220, 560, 260],
      "confidence": 0.97,
      "text": "Total: NPR 24,500"
    }
  ]
}

Set bbox to true when you need word-level coordinates for highlighting.

Errors

Status	Code	Fix
`400`	`missing_artifact`	Attach a valid `document` file to the request.
`400`	`unsupported_format`	Use PDF, PNG, JPG, TIFF, or WEBP only.
`401`	`unauthorized`	Re-export a valid `SAGEA_API_KEY` value.
`413`	`payload_too_large`	Compress or split file under 100MB.
`429`	`rate_limit_exceeded`	Back off and retry with exponential delay.

Limits

ARVA OCR allows 30 requests per minute on Starter and 300 requests per minute on Pro.