Data Control
Exploring ways to ensure only necessary data is shared with AI models.
What's This About?
LLMs are powerful but they can leak data, get tricked by prompts, or be misused in code.
GateLLM is a secure proxy layer for LLMs. It's like a firewall but for prompts and AI output.
terminal ~ zsh
$ npm install gatellm
Installing GateLLM secure proxy...
Starting GateLLM proxy server on port 3000...
✓ Connected to OpenAI API
✓ Connected to Anthropic API
✓ Local model loaded: mistral-7b
◯ Securing requests with PII filtering
◯ Prompt injection protection active
GateLLM ready! All AI traffic is now secure.
█
We're building:
- Prompt validation & injection detection
- Logging & auditing for LLM requests
- Guardrails for AI-assisted code review
- A developer-friendly gateway with real-world use cases
Why GateLLM?
Secure Access
Working on a single interface for managing multiple LLM integrations securely.
Designed for Privacy
Starting from scratch to build something robust for those who need more control.
Built by Developers, for Developers
This isn't a product, it's a journey.
If you're a:
- • CS student learning AI & security,
- • Junior dev exploring prompt engineering,
- • Senior engineer building LLM features into your app
Come help shape GateLLM ~ we're sharing everything as we learn.
Proof of Concept
Starting Small, Dreaming Big
As a developer, I'm building this project step-by-step, starting with a simple command-line interface that can:
- Detect PII in prompts before they reach AI models
- Route requests to different LLM providers
- Log and analyze prompts for security concerns
gatellm-cli.py
import argparse
import re
# Simple PII detector
def detect_pii(text):
# Match emails, phone numbers, etc.
patterns = [r'[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}']
for pattern in patterns:
if re.search(pattern, text):
return True
return False
def main():
# CLI is under development
# More code coming soon...
Next Steps:
Phase 1: Simple CLI tool for prompt scanning (NLP basics)
Phase 2: Basic REST API
Technical Exploration
Prompt Injection Protection
Rule-based detection to block harmful prompts
PII Detection & Filtering
Protect sensitive data from reaching AI models
Authentication System
API keys for secured, controlled access
Logging & Auditing
Track all requests for security analysis
Early Explorers Wanted
We're looking for early adopters to test the CLI, try prompt firewalls, and co-create best practices for LLM dev.
Be part of the conversation. Help define the future of secure AI development.
- Add your name to get early access
- Feedback makes this better for everyone