🧾 easy-e-rechnung

Java-App for creating and validating Factur-X / ZuGFeRD / X-Rechnung invoices conforming with EU-Norm EN 16931.

---

✨ Key Features

Feature	Description
🇪🇺 EU Compliant	Generates invoices conforming to EN 16931, accepted across all EU member states.
🔒 100% Offline & Private	All processing happens locally on your machine. Your invoice data never leaves your computer.
🤖 LocalAI-Powered OCR	Uses local, open-weight AI models for automatic, high-quality text extraction from PDF invoices.
📄 Multi-Page Support	Processes multi-page PDF invoices — each page is OCR'd separately, then semantically merged by the LLM into a single structured result.
🧠 Works with Ollama	Integrates with Ollama for local LLM inference. Default models: glm-ocr:q8_0 (OCR) and qwen3:4b-q8_0 (JSON extraction).
💻 Cross-Platform	Runs on macOS, Linux, and Windows. (macOS and Linux are the primary tested platforms.)

🚀 Setup

# Install tooling and dependencies
bash setup.sh

🧾 Usage

Users usually interact with the app via the GUI:

# Start the App (Java)
./start.sh

🧑‍💻 Calling the OCR pipeline via Shell

You can also run the OCR pipeline directly via shell:

# Single-page PDF
bun run src/ocr.ts --input demo/verify.pdf --output /tmp/result.json \
  --seller-address "Friedrich-Damm-Str. 8, 80999 München" \
  --seller-tax-no "147/214/00001"

# Multi-page PDF with custom models
bun run src/ocr.ts --input demo/verify_multipage.pdf --output /tmp/result.json \
  --seller-address "Friedrich-Damm-Str. 8, 80999 München" \
  --seller-tax-no "147/214/00001" \
  --ocr-model glm-ocr:q8_0 --json-model qwen3:4b-q8_0

OCR Pipeline Architecture

1. PDF → Images — Each page is rendered as a high-resolution image.
2. Image Preprocessing — Contrast boost, normalization, and resize to max 3MP.
3. OCR per Page — Vision model (glm-ocr:q8_0) extracts text as markdown.
4. Date Preprocessing — Date ranges in the OCR text are annotated with day counts (e.g. DAYS: 31) to help the LLM correctly set quantities for time-based line items.
5. JSON Extraction — All page markdowns are combined and sent in a single LLM call (qwen3:4b-q8_0) to produce a structured ZUGFeRD-compatible JSON.

---

📸 How It Works

Step 1: Drag & Drop Your Invoice PDF

Simply drag and drop a PDF invoice into the app. Multi-page PDFs are displayed with tabs on the left side. The AI-powered OCR will automatically extract the text from each page.

---

Step 2: AI Post-Processing

The local AI model analyzes the OCR output and intelligently extracts all relevant invoice data. Progress is shown in real-time with per-page OCR status tabs and a JSON extraction log.

---

Step 3: Review Invoice Positions

Review and edit the extracted line items. The app calculates totals automatically.

---

Step 4: Review Taxes & Totals

Verify the tax calculations and monetary summations.

---

Step 5: Create the e-Invoice

Click to generate the ZuGFeRD/Factur-X compliant PDF with embedded XML.

---

Step 6: Invoice Created Successfully

The app confirms successful creation and automatically opens the ELSTER e-Rechnung portal for official validation.

---

Step 7: View the Final Output

Your new e-Invoice is ready.

---

✅ Validated by Official Tools

ELSTER (German Tax Authority)

Other Validators

---

📁 Demo Data

The demo/ folder contains sample invoice data for testing and demonstration purposes.

---

📜 License

MIT, Open Source.

---

🛡️ Privacy Promise

• No network requests. All AI inference runs locally.
• No telemetry. Your data stays on your device.
• Open Source. Audit the code yourself.