Structured data from your documents.
Create a project, drop in invoices, plane tickets, training documents or purchase orders, and get back a strict JSON schema ready to ship to your ERP.
Supported document types
Invoice
FactureSupplier, customer, line items, VAT breakdown, totals, payment terms. Priority doc type.
Plane Ticket
Billet d'avionPNR, ticket number, passenger, flight segments, fare breakdown.
Training Document
Document de formationTraining org, trainee, programme, durations, Qualiopi / RNCP refs (French training).
Purchase Order
Bon de commandeBuyer, supplier, order lines, delivery + payment terms, totals.
Models
Donut
CORD-v2 receipt / invoice parser. Input is just an image; output is a CORD-schema dict we remap to the invoice shape. Ignores prompts.
replicate: willywongi/donut
Claude 4 Sonnet
Anthropic Claude 4 Sonnet via Replicate. General multimodal model; we embed the exact target JSON schema in the prompt and coerce the reply with Zod.
replicate: anthropic/claude-4-sonnet
DeepSeek-OCR (text only)
High-accuracy OCR that returns plain markdown / text. Does NOT produce structured fields — use Donut or Claude for the table view.
replicate: lucataco/deepseek-ocr