OCR

Mistral OCR extracts text and document structure from scanned PDFs and images. It is a separate provider model, not an input mode on completionModel(...).

import { MistralClient } from '@anvia/mistral'

const mistral = new MistralClient({
  apiKey: process.env.MISTRAL_API_KEY,
})

const ocr = mistral.ocrModel()

Process a document URL

const result = await ocr.ocr({
  source: {
    type: 'document_url',
    url: signedDocumentUrl,
    documentName: 'invoice.pdf',
  },
  tableFormat: 'markdown',
  includeImageBase64: false,
})

console.log(result.markdown)

The URL must be reachable by the provider for the duration of the request. Prefer a short-lived signed URL over making a private document permanently public.

Process local bytes

import { readFile } from 'node:fs/promises'

const result = await ocr.ocr({
  source: {
    type: 'bytes',
    data: await readFile('./invoice.pdf'),
    filename: 'invoice.pdf',
    expiry: 3600,
    visibility: 'user',
  },
  pages: [0, 1],
  includeImageBase64: false,
})

For byte sources, the adapter uploads the file to Mistral with the ocr purpose and then processes its returned file ID. result.uploadedFile exposes normalized upload metadata. The adapter does not make provider retention decisions for the application, so configure expiry and cleanup according to the document's sensitivity.

OCR also accepts an image_url or an existing Mistral file_id source. Reject empty byte sources and empty filenames before they reach the provider.

Use the normalized result

Field	Meaning
text / markdown	Combined page Markdown, suitable for display or the next processing step.
pages	Normalized page entries with indexes, Markdown, images, and available structural metadata.
documentAnnotation	Optional provider annotation when requested.
usageInfo	Provider usage details when returned.
uploadedFile	Upload metadata for a byte source.
rawResponse	Original provider response for narrow diagnostics.

Preserve page boundaries when downstream citations or review need provenance:

const pages = result.pages.map((page) => ({
  id: `${documentId}#page=${page.index}`,
  text: page.markdown,
  pageIndex: page.index,
}))

Leave includeImageBase64 disabled unless a later step actually needs extracted images. Base64 images can substantially increase response, memory, and trace sizes.

Validate before use

OCR output can contain recognition errors. Validate totals, dates, identifiers, and other critical values before product writes, and never treat OCR as proof that a document is authentic or that a user may access it.

For deciding when to use OCR, loaders, or model-visible media, continue to the broader OCR guide.