privylayer 0.2.5

Privacy layer for AI prompts

🔐 PrivyLayer

Privacy layer for AI systems — protect sensitive data before it reaches the LLM.

---

🚀 Overview

PrivyLayer is a lightweight Python library designed to act as a secure buffer between user input and AI models. It detects sensitive information (PII), replaces it with secure tokens, and restores the original data after the AI response—ensuring zero data exposure to third-party providers.

---

⚡ Why PrivyLayer?

Modern AI systems often process raw user data, which may contain:

• Identity: Aadhaar, PAN
• Contact: Phone numbers, Emails
• Finance: Bank details, IFSC codes

PrivyLayer solves this by introducing a reversible masking system, allowing AI to understand context without ever seeing the actual private values.

---

🧠 How It Works

1. User Input: "My Aadhaar is 1234 5678 1234."
2. Mask (PrivyLayer): "My Aadhaar is ⟦PII_AADHAAR_x1⟧." (Validates checksum internally).
3. AI Model: Processes the masked text.
4. Restore (PrivyLayer): Swaps the token back for the original value.
5. Final Output: Securely displayed to the user.

---

✨ Features

• 🔍 Advanced Detection: Uses optimized patterns for PII.
• ✅ Aadhaar Validation: Now includes checksum verification to notify if an Aadhaar number is real or fake.
• 🔁 Reversible Masking: Maintain full context without data loss.
• 🧠 Smart Tokenization: Structured ⟦PII_TYPE_ID⟧ format for easy LLM processing.
• ⚡ Fast & Lightweight: Minimal overhead for real-time applications.
• 🔒 Privacy First: No data is stored or sent to any external server.

---

📦 Installation

pip install privylayer

---

🛠️ Usage

🔐 Protect (Mask data)

import privylayer

text = "My phone is 9876543210 and Aadhaar is 987654321096"
# find_p analyzes the text and returns masked version, context, and metadata
mask, context, seen = privylayer.mask(text)

print(mask)

🔓 Restore (Unmask data)

from privylayer import restore

original = restore(masked_response, context)
print(original)
# Output: My phone is 9876543210 and Aadhaar is [Aadhaar Redacted]

---

🧩 Supported Data Types (India-Specific)

• 📱 Phone numbers
• 📧 Email addresses
• 🪪 Aadhaar numbers (With Checksum Validation)
• 🧾 PAN numbers
• 🏦 IFSC codes

---

🛡️ Aadhaar Checksum Validation

The latest version of PrivyLayer automatically validates the integrity of Aadhaar numbers using the Verhoeff algorithm.

• Note: The library will print a system message indicating whether the detected Aadhaar is Real or Fake based on the checksum, ensuring higher data quality before masking.

---

🚀 Roadmap

• AI-based entity and name detection
• Risk scoring system for data sensitivity
• Open for all over the world for major countries with specfic detection
• API Gateway integration

---

⚠️ Limitations

While PrivyLayer provides a robust security buffer, users should be aware of the following technical boundaries:

No Official PAN/IFSC Verification: The library detects PAN cards and IFSC codes using structural pattern matching (Regex). It does not connect to government databases or banking APIs to verify if the identity or bank branch actually exists.

Aadhaar Checksum Only: For Aadhaar, the library performs a Verhoeff algorithm checksum validation. This confirms if the number is mathematically valid (Real vs. Fake format), but it does not perform biometric or OTP-based authentication.

Regex-Based Detection: Since the core engine relies on sophisticated regular expressions, it may occasionally struggle with high-accuracy name detection if the names are common dictionary words.

Context Dependency: The restore() function requires the original context object generated during the protect() phase. If this context is lost or corrupted, the masked data cannot be recovered.

No Global Storage: PrivyLayer is entirely stateless. While this is a security feature, it means you are responsible for securely managing the mapping context within your own application environment. We always look forward to provide Ultimate Privacy

---

📄 License

MIT License

👨‍💻 Author

Built by Nithish 🚀 Portfolio | GitHub

---

🌟 Vision: Making AI safe for everyone by protecting sensitive data at the source.