Skip to main content

German-Privacy-Shield

DSGVO-konforme Dokument-Redaktion mit KI

Erkennt und schwärzt personenbezogene Daten (PII) in deutschen Dokumenten automatisch. Läuft 100% lokal mit Ollama - keine Cloud, keine Datenübertragung.

Highlights

  • 100% Lokal - Läuft komplett auf Ihrer Hardware mit Ollama
  • DSGVO-konform - Basiert auf Art. 4 DSGVO, Art. 6 DSGVO, Par. 26 BDSG
  • Smart Knowledge Injection - DSGVO-Wissen direkt im Modell integriert
  • Multi-Mode - Drei Sicherheitsstufen (strict, balanced, loose)
  • Dokumenttyp-Erkennung - Optimiert für Rechnungen, Verträge, Ausweise, etc.
  • Nur 1.9GB - Läuft auf jeder Consumer-GPU und CPUs

Installation

pip install german-privacy-shield

Voraussetzungen

  1. Ollama installieren: https://ollama.com
  2. Modell laden:
ollama pull Keyvan/german-privacy-shield

Schnellstart

Python API

from german_privacy_shield import GermanPrivacyShield, Redactor, PrivacyMode

# PII-Erkennung
shield = GermanPrivacyShield(mode=PrivacyMode.BALANCED)
result = shield.detect_pii("rechnung.png")

for pii in result.pii_items:
    print(f"[{pii.label}] {pii.text}")

# Dokument schwärzen
redactor = Redactor(mode=PrivacyMode.STRICT)
result = redactor.redact("rechnung.png", "rechnung_redacted.png")
print(f"Geschwärzt: {result.pii_count} PIIs")

CLI

# PII-Erkennung
german-privacy-shield detect rechnung.png

# Dokument schwärzen
german-privacy-shield redact rechnung.png -o rechnung_redacted.png

# Mit strenger Sicherheitsstufe
german-privacy-shield redact dokument.png --mode strict

# Batch-Verarbeitung
german-privacy-shield batch ./input/ ./output/

Sicherheitsstufen

Modus Beschreibung Anwendungsfall
strict Alle PIIs, Unterschriften, Fotos Maximaler Datenschutz, AI-Training
balanced Namen, Adressen, Bankdaten Standard DSGVO-Compliance
loose Nur IBAN, Steuer-ID, Sozialvers. Minimale Schwärzung

Erkannte PII-Kategorien

  • Identität: Namen, Geburtsdaten, Ausweisnummern
  • Kontakt: Adressen, Telefon, E-Mail
  • Finanzen: IBAN, BIC, Kontonummern, Beträge
  • Biometrie: Unterschriften, Fotos, Gesichter
  • Gesundheit: Diagnosen, Versicherungsnummern
  • Arbeit: Personalnummern, Gehalt
  • Steuern: Steuernummern, Steuer-ID

Dokumenttypen

Das Modell ist für folgende Dokumenttypen optimiert:

  • invoice - Rechnungen
  • contract - Verträge
  • id_card - Ausweise und Reisepässe
  • tax - Steuerbescheide
  • medical - Arztbriefe und medizinische Dokumente
from german_privacy_shield import Redactor, DocumentType

redactor = Redactor()
result = redactor.redact("rechnung.png", doc_type=DocumentType.INVOICE)

API Referenz

GermanPrivacyShield

class GermanPrivacyShield:
    def __init__(
        self,
        model: str = "german-privacy-shield",
        mode: PrivacyMode = PrivacyMode.BALANCED,
        host: Optional[str] = None  # Ollama Host URL
    )

    def detect_pii(
        self,
        image_path: str,
        doc_type: Optional[DocumentType] = None,
        custom_rules: Optional[List[str]] = None
    ) -> RedactionResult

Redactor

class Redactor(GermanPrivacyShield):
    def redact(
        self,
        image_path: str,
        output_path: Optional[str] = None,
        doc_type: Optional[DocumentType] = None,
        custom_rules: Optional[List[str]] = None,
        draw_labels: bool = False
    ) -> RedactionResult

    def redact_batch(
        self,
        input_dir: str,
        output_dir: str,
        pattern: str = "*.png"
    ) -> List[RedactionResult]

Rechtlicher Hintergrund

German-Privacy-Shield basiert auf:

  • Art. 4 DSGVO - Definition personenbezogener Daten
  • Art. 6 DSGVO - Rechtmäßigkeit der Verarbeitung
  • Par. 26 BDSG - Pseudonymisierung und Anonymisierung
  • Par. 22 BDSG - Besondere Kategorien (Gesundheit, Religion)

Performance

Hardware Zeit pro Dokument
RTX 4060 8GB ~5 Sekunden
RTX 3060 12GB ~6 Sekunden
CPU (keine GPU) ~15 Sekunden

Systemanforderungen

  • Python 3.9+
  • Ollama v0.1.0+
  • 4GB+ VRAM (GPU) oder 8GB+ RAM (CPU)
  • 2GB Speicherplatz für Modell

Verwandte Projekte

Lizenz

Apache 2.0

Autor

Keyvan Hardani - keyvan.ai


Made with privacy in Germany

Metadata

Release files for german-privacy-shield 0.1.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for german-privacy-shield 0.1.0
File Size Uploaded
german_privacy_shield-0.1.0.tar.gz 13.0 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for german-privacy-shield 0.1.0
File Interpreter ABI Platform
german_privacy_shield-0.1.0-py3-none-any.whl Python 3 none any Details

Total release size: 25.9 kB

Release files / german_privacy_shield-0.1.0.tar.gz

Download URL german_privacy_shield-0.1.0.tar.gz
Size 13.0 kB
Tags Source
SHA-256 checksum
How to use checksums
cf2e7f239f9a22ae01ffd41d35d4a494524d0ce6174ef452d4cf5fdb28d382c6
BLAKE2b-256 checksum
How to use checksums
c2833f1ef725c4ba43019b10eb084663b012708651c6a28d00b69885a2ac2b25
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.7

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Dec 7, 2025.

Transparency log

Release files / german_privacy_shield-0.1.0-py3-none-any.whl

Download URL german_privacy_shield-0.1.0-py3-none-any.whl
Size 12.9 kB
Tags Python 3
SHA-256 checksum
How to use checksums
536716709b05601206176117cbf81af81d0effbc8244215f33055bafab4ff492
BLAKE2b-256 checksum
How to use checksums
97e842184dd3d3f43195e6fc98614e7e2b1c828967889d8b09834a97ab6e0589
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.7

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Dec 7, 2025.

Transparency log

Release history Release notifications | RSS feed

This release

0.1.0 This release

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page