⚡ Overview

Enterprise PII Guardrails Studio is an ultra-fast, defense-in-depth privacy gateway designed to intercept, detect, mask, and pseudonymize Personally Identifiable Information (PII) before it ever touches public LLM APIs (OpenAI, Anthropic, Gemini) or third-party vector databases.

Equipped with an in-memory authorization cache, asynchronous audit pipeline, and multi-tier regex + NLP models, it sustains sub-25ms live masking latency while safeguarding compliance with GDPR, HIPAA, PCI-DSS, and CCPA.

🚀 Quickstart

Method 1: Using pip (Standard Python)

# 1. Install from PyPI
pip install piiguardrails

# 2. Launch the studio
piiguardrails

Method 2: Using uv (Instant Sandbox — Zero Setup)

Run directly in a self-contained, isolated environment without polluting global Python:

uv run --with piiguardrails piiguardrails

🖥️ First-Run Setup & UI Access

On first launch, piiguardrails automatically provisions the core engine and configures your encrypted local database.

1. A quick 30-second setup wizard prompts you in the terminal for your Company Name and Admin Password.
2. Once initialized, the studio launches on: 👉 http://localhost:8000
3. Log in with your admin credentials to access the live dashboard, manage API keys, customize PII detection rules, and monitor telemetry.

🧩 Python Integration Example

Once the studio is running, integrate it directly into your LangChain, LlamaIndex, or raw API pipeline:

import requests

API_KEY = "your-api-key-here"  # Generated in the Studio UI
headers = {"X-API-Key": API_KEY, "Content-Type": "application/json"}

# 1. Mask sensitive input before sending to LLM
payload = {
    "text": "Hello Alice Smith, your verification code was sent to alice.smith@acme.corp. Call us at 415-555-2671."
}

mask_response = requests.post("http://localhost:8000/mask", json=payload, headers=headers).json()
print("Masked text for LLM:", mask_response["masked_text"])
# Output: "Hello <PERSON_1>, your verification code was sent to <EMAIL_1>. Call us at <PHONE_1>."

# 2. Query your LLM with safe, masked prompt...
# simulated_llm_reply = "We contacted <PERSON_1> at <EMAIL_1> regarding the ticket."

# 3. Unmask the LLM response back to real entities for your user
unmask_payload = {
    "text": "We contacted <PERSON_1> at <EMAIL_1>.",
    "mapping": mask_response["mapping"]
}
unmask_response = requests.post("http://localhost:8000/unmask", json=unmask_payload, headers=headers).json()
print("Restored response:", unmask_response["unmasked_text"])
# Output: "We contacted Alice Smith at alice.smith@acme.corp."

🛡️ Key Capabilities & Architecture

📋 System Requirements

• Operating System: Windows 10, Windows 11, Windows Server (64-bit)
• Python: >= 3.8 (or uv)
• Memory: Minimum 2 GB RAM
• Port: Default 8000 (configurable via .env or --port)

📄 License

Proprietary / Enterprise Commercial License. All rights reserved.