99% extraction accuracy on German B2B invoices

Turn invoices into
structured data

Upload any German B2B invoice — PDF, scan, or ZUGFeRD — and get precise, machine-readable JSON in seconds. No templates, no rules, just intelligence.

invoice_2024_0471.pdf
Extracted — 28 fields
{
  "invoice_number": "471102",
  "invoice_date": "05.03.2024",
  "sender_name": "Lieferant GmbH",
  "total_amount": 235.62,
  "vat_amount": 37.62,
  "iban": "DE89 3704 0044 ...",
  "items": [
    { "description": "Trennblätter A4",
      "quantity": 20,
      "unit_price": 9.90 }
  ]
}

99%

Extraction accuracy

Macro F1 on 4 000+ invoices

50+

Fields extracted

Dates, amounts, line items, IBANs

<6s

Per invoice

GPU-accelerated inference

5

Invoice domains

Energy, health, gov, B2B, more

Features

Built for German invoice complexity

Five invoice domains. Fifty extraction fields. One API call.

Vision AI Extraction

State-of-the-art vision-language model reads invoices like a human — no templates or rule engines needed.

ZUGFeRD & XRechnung

Native support for e-invoicing standards. Embedded XML is parsed directly; visual fallback for scanned PDFs.

5-Second Turnaround

GPU-accelerated inference returns structured JSON in seconds, not minutes. Process thousands of invoices per hour.

Dedicated GPU Infrastructure

Your invoices are processed on our own GPU servers — not routed through shared multi-tenant AI services.

50+ Fields Extracted

From invoice number and dates to line items, IBAN, VAT IDs, meter readings — everything structured automatically.

5 Invoice Domains

Energy, healthcare, government, standard B2B, and more. Domain-specific models for maximum accuracy.

Trust & Sovereignty

Your data stays yours

Built for European compliance. Self-hostable. No vendor lock-in.

No Third-Party AI

Your invoices never leave our infrastructure. We run our own vision model on dedicated GPUs — no OpenAI, no external APIs, no data sharing.

Self-Hosting Available

Enterprise customers can deploy Documation on-premise in their own Rechenzentrum. Full control over hardware and data lifecycle.

Delete Anytime

Full control over your data. Delete individual extractions or your entire account — all stored files and results are permanently removed.

Open-Weight Model

Built on open-weight AI, not a proprietary black box. No vendor lock-in — you can inspect, audit, and self-host the underlying model.

Encrypted at Rest & Transit

All connections are TLS-encrypted. Stored documents and results are protected with AES-256 encryption via our database provider.

EU Data Residency Roadmap

Currently hosted on dedicated GPU infrastructure. EU-based Rechenzentrum deployment planned for enterprise customers.

How It Works

Three steps to structured data

01

Upload your invoice

Drag and drop a PDF, scanned image, or ZUGFeRD file. We accept any German B2B invoice format.

02

AI extracts the data

Our vision model reads the invoice end-to-end — headers, line items, footers — and maps every field to structured JSON.

03

Get structured results

Download clean JSON or push to your ERP via API. Review flagged fields in the human-in-the-loop dashboard.

Pricing

Simple, transparent pricing

Start free. Pay per page when you scale.

Trial

Free

Test with real invoices — no credit card required.

  • 20 pages / day
  • All 5 invoice domains
  • JSON export + download
  • Dashboard history
Start free trial
Most popular

Pro

€0.05/ page

Credit-based pricing that scales with your volume.

  • Unlimited pages
  • Priority GPU queue
  • API access + webhooks
  • HITL review dashboard
  • Dedicated support
Coming soon

Enterprise

Custom

On-prem deployment in your own Rechenzentrum.

  • Self-hosted option
  • Custom model fine-tuning
  • SSO via Zitadel
  • SLA guarantee
  • Dedicated account manager
Contact sales

Ready to automate your invoice processing?

Upload your first invoice in 30 seconds. No credit card, no setup, no strings attached.