Privacy proxy for LLMs
Open Source

Mask personal data before it reaches AI.

Names, emails, and secrets are automatically replaced with placeholders. Works with OpenAI, Azure, and any OpenAI-compatible API.

20+ PII types 24 languages Auto-unmask responses Self-hosted
ChatGPT
AI receives: ‹EMAIL› ‹PERSON›
Auto-unmasked
Message ChatGPT...

Two ways to protect your data

Choose your protection level: mask sensitive data before sending to your provider, or route private requests to a local LLM.

Default

Mask Mode

Replaces personal data with placeholders before sending to your LLM provider. Responses are automatically unmasked.

Your app sends:
Email to Dr. Sarah Chen
OpenAI receives:
Email to <PERSON_1>
Your app receives:
Dear Dr. Sarah Chen...
  • No local LLM needed
  • Same model quality for all requests
  • Set up in minutes
Advanced

Route Mode

Requests containing personal data go to your local LLM. Everything else goes to your configured provider.

"Help with code review"
OpenAI (cloud)
"Email to Dr. Sarah Chen..."
Ollama (local)
  • PII never touches external servers
  • Cloud quality + local privacy combined
  • Use your existing Ollama or vLLM

What we protect

PasteGuard detects 20+ types of sensitive information across 24 languages.

Names

Dr. Sarah Chen, John Smith

Emails

sarah.chen@hospital.org

Phone Numbers

+1 555 123 4567

Credit Cards

4111-1111-1111-1111

Locations

New York, Berlin

IBANs

DE89 3704 0044...

IP Addresses

192.168.1.1

API Keys & Secrets

sk-proj-..., ghp_...

Private Keys

OpenSSH, PEM, RSA keys

JWT Tokens

Bearer tokens, sessions

PII detection powered by Microsoft Presidio • Supports 24 languages including English, German, French, Spanish, Italian, Chinese, Japanese, and more

How it works

PasteGuard sits between your app and your LLM provider. Just change the API endpoint.

1

You send a request

Your app sends requests to PasteGuard's OpenAI-compatible API endpoint.

POST /openai/v1/chat/completions
2

PasteGuard protects

PII and secrets are detected and masked or routed based on your configuration.

Dr. Sarah Chen <PERSON_1>
3

Response auto-unmasked

Responses are automatically unmasked so you see the original values—even in streaming.

You see: Dr. Sarah Chen

Quick Start

Get PasteGuard running in minutes with Docker. Point your app to localhost and you're protected.

Terminal
# Clone and start
git clone https://github.com/sgasser/pasteguard.git
cd pasteguard
docker compose up -d
# Point your app to PasteGuard
OPENAI_BASE_URL=http://localhost:3000/openai/v1
Coming Soon
Free & Open Source

For ChatGPT, Claude & Gemini users

Paste customer data into AI chats without the risk. The browser extension masks sensitive information before it leaves your browser. Works on Chrome, Firefox, Edge, and Safari.

  • Auto-mask on paste

    PII is masked the moment you paste

  • Visual unmasking

    See original values while AI only sees placeholders

  • Works everywhere

    ChatGPT, Claude, Gemini, Copilot, Perplexity

Join the Beta

Join our early tester community. Your direct feedback helps us ensure quality and build the right features.

Join the waitlist

No spam. We'll email when the extension is ready.

PasteGuard
Active • Protected
This session
0
Items masked
0%
Protected
Emails 5 masked
Names 4 masked
Phone numbers 3 masked

Open Source. Auditable. Trustworthy.

PasteGuard is fully open source under the Apache 2.0 license. Audit the code, self-host it, or contribute to make it better.

Apache 2.0
Open Source License
0
Languages Supported
0
PII Types Detected
Star on GitHub