Document Intelligence

Read the page.
Keep the context.

Document models cover the steps before and after extraction: read page structure, separate mixed packets, and answer questions with references to the original document.

Input

PDFs or images, a document taxonomy, or a question.

Output

Page-aware content, component document ranges, or cited answers.

Evaluation

Reading order, split boundaries, citation accuracy, and answer support.

document.parse

Parse

Convert a PDF or image into page-aware content that downstream models can use.

Inspect sample
Sample output
{
  "pages": [
    {
      "page": 1,
      "blocks": [
        {
          "kind": "heading",
          "text": "Invoice INV-204"
        },
        {
          "kind": "text",
          "text": "Total due USD 480.00"
        }
      ]
    }
  ]
}
document.split

Split

Find component document boundaries and return their page ranges.

Inspect sample
Sample output
{
  "documents": [
    {
      "type": "cover_sheet",
      "pages": [
        1
      ]
    },
    {
      "type": "invoice",
      "pages": [
        2,
        3
      ]
    },
    {
      "type": "packing_list",
      "pages": [
        4
      ]
    }
  ]
}
document.ask

Grounded Q&A

Ask about a document and inspect the passage that supports the answer.

Inspect sample
Sample output
{
  "answer": "Within 30 days of the invoice date.",
  "citations": [
    {
      "page": 3,
      "text": "Payment is due within 30 days of the invoice date."
    }
  ]
}
Model availability

This research interface is not yet available through a public API. Explore the examples or contact us about your application.

Build what comes next.

Talk to Istren