DOCUMENT INTELLIGENCE

Move from OCR output to usable document data.

Build on a common document layer, then choose parsing, extraction, or workflow-ready output for the job at hand.

Document IR
Keep providers behind one normalized representation
Schema extraction
Ask for the fields your application needs
Confidence-aware results
Route uncertain fields to the right review step

CHOOSE THE PRIMITIVE

Start with the result your workflow needs.

Use OCR and parsing for document understanding, then apply structured extraction when a system needs fields it can act on.

OCR API
Text, blocks and confidence
PDF to Markdown
LLM-ready structured Markdown
Document to JSON
JSON fields that match your schema
Table Extraction
Rows, cells and CSV
Invoice Extraction
Normalized AP data

ONE DOCUMENT PIPELINE

Keep the next action explicit.

Document → parse → extract → validate → workflow. Keep the same source context through every handoff.

Spreadsheet data
Table Extraction
Vendor invoices
Invoice Extraction
A complete document result
Markdown + JSON

ONE DOCUMENT LAYER

Keep models replaceable, not your product logic.

Normalize once, then let your application depend on a stable document contract instead of a provider-specific response.

DOCUMENT → DOCUMENT IR → OUTPUT
provider · normalized structure · schema · confidence · workflow

START WITH A REAL DOCUMENT

Inspect the result before you integrate.

Use the Playground to compare a source document with the data your system receives.