VectorUnforget 🛡️ Vector DB Right-to-be-Forgotten Engine

VectorUnforget is an enterprise-grade Python engine designed to enforce GDPR/CCPA compliance (Right to be Forgotten) across Vector Databases used in RAG (Retrieval-Augmented Generation) architectures.

🚀 Key Features

• Multi-Vector DB Support (Adapter Pattern): Out-of-the-box integration with ChromaDB, Qdrant, and Pgvector (PostgreSQL), with an extensible architecture for enterprise vector stores.
• Global PII Recognition (NER & Regex): Identifies primary entries and automatically extracts secondary PII globally:
  • Universal: Emails, International Phone Numbers, Credit Cards, IBANs, IPv4/IPv6 Addresses.
  • National Identifiers: US SSN, UK NINO, Italian Fiscal Code, Canadian SIN, German Steuer-ID.
  • Multilingual NER: Powered by spaCy for custom entities (PERSON, ORG, GPE, FAC).
• Dry Run Mode (Simulation): Preview vectors and secondary PII targeted for removal before committing destructive deletes to production databases.
• Cascading PII Erasure: Automatically purges orphaned vector entries that share secondary PII even if the target name is not explicitly mentioned.
• Name Variant Engine: Automatically generates and matches name permutations (e.g., Mario Rossi, M. Rossi, Rossi M.).
• Tamper-Proof Audit Trail: Generates a SHA-256 signed JSON certificate of erasure for compliance auditors and DPOs.

📦 Installation & Setup

1. Install via pip (with optional extra dependencies):

   # Install core package
   pip install vector-unforget
   
   # Install with specific adapters (e.g., Qdrant or Pgvector)
   pip install "vector-unforget[qdrant]"
   pip install "vector-unforget[pgvector]"
   
   # Or install all adapters:
   pip install "vector-unforget[all]"

2. Download spaCy model for NER support:

   python -m spacy download en_core_web_sm

🛠️ Usage Examples

1. Global Dry Run (Simulation Mode)

from qdrant_client import QdrantClient
from vector_unforget import VectorUnforgetEngine, QdrantAdapter

client = QdrantClient("http://localhost:6333")
adapter = QdrantAdapter(client=client, collection_name="production_rag")
engine = VectorUnforgetEngine(adapter=adapter, db_name="qdrant_prod")

# Run simulation detecting US SSN, IPv4, Emails, and phones globally
preview = engine.purge_user("John Smith", dry_run=True)
print("Vectors to be purged:", preview["vector_ids_to_be_purged"])
print("Secondary PII extracted:", preview["secondary_pii_extracted"])

2. Pgvector (PostgreSQL) Integration

from vector_unforget import VectorUnforgetEngine, PgvectorAdapter

adapter = PgvectorAdapter(
    connection_string="postgresql://user:password@localhost:5432/rag_db",
    table_name="embeddings",
    id_column="id",
    text_column="content"
)

engine = VectorUnforgetEngine(adapter=adapter, db_name="pgvector_prod")

# Execute real cascading purge
audit_log = engine.purge_user("John Smith", dry_run=False)

📄 License

This project is licensed under the GNU Affero General Public License v3.0 (AGPLv3) - see the LICENSE file for details. Commercial licensing options are available for enterprise integration.