Overview

An intelligent text anonymization and de-anonymization system that allows users to safely interact with AI chatbots while protecting sensitive information. Built on Microsoft Presidio for enterprise-grade entity recognition, this tool provides a seamless workflow for redacting, processing, and restoring confidential data.

Key Features

Intelligent Recognition

Automatically detects and anonymizes emails, organizations, URLs, UN entities, department codes, and custom patterns you define.

AI-Powered Patterns

Uses LLMs to generate regex patterns from natural language. Just describe what you want to match, and the AI creates the pattern.

Reversible Process

Maintains mapping to restore original values. Each unique entity gets a consistent, distinct placeholder for traceability.

Integrated Chat

Chat with AI models using anonymized text. Messages are automatically redacted before sending and restored after receiving responses.

Custom Patterns

Add entity patterns dynamically via web interface. Patterns persist and support priority scoring for overlapping matches.

Privacy-First

All anonymization happens locally. Use OpenAI or local Ollama models for complete data privacy and control.

Screenshots

Screenshot 1: (De)Anonymization interface

Screenshot 2: (De)Anonymization & Chat Interface

How It Works

1. Detect
Presidio identifies sensitive entities

→

2. Anonymize
Replace with fake values

→

3. Process
Send to AI safely

→

4. Restore
De-anonymize response

The system uses regex-based entity detection with configurable scoring, overlap detection, and conflict resolution. Case-insensitive matching with word boundaries and negative lookaheads ensures accurate recognition while excluding similar words.

Use Cases

Research Organizations: Share drafts with AI assistants without exposing institutional names
Legal/Compliance: Review documents with AI while redacting sensitive entity information
Academia: Get writing assistance while protecting university and department identities
Data Analysis: Analyze sensitive datasets with AI while maintaining privacy
General Privacy: Chat with AI about sensitive topics with automatic redaction

Technical Stack

Flask + Python Microsoft Presidio LangChain OpenAI API Ollama Faker RESTful API Regex Engine

Configuration Options

AI Provider: Choose between OpenAI or local Ollama for complete privacy
Models: Configure which model to use (GPT-4, Gemma, Llama, etc.)
Custom Patterns: Add unlimited entity types with custom regex patterns
Scoring Priority: Control which patterns take precedence when overlapping
Deployment: Local development mode or Docker containerization