AIProxyGuard

LLM Security Proxy with Prompt Injection Detection.

What It Does

AIProxyGuard sits between your application and LLM providers to detect and block malicious inputs before they reach the model. Point your OpenAI/Anthropic SDK at the proxy instead of directly at the provider.

Quick Start

# Run the proxy
docker run -d -p 8080:8080 ghcr.io/ainvirion/aiproxyguard:latest

# Verify it's running
curl http://localhost:8080/healthz

Point your LLM client to the proxy:

from openai import OpenAI

client = OpenAI(
    api_key="sk-...",
    base_url="http://localhost:8080/openai/v1"
)

# Normal requests work as expected
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

# Malicious requests are blocked
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Ignore all previous instructions..."}]
)
# Raises: BadRequestError - content_blocked

Detection-Only Mode

Use the /check endpoint to scan text without forwarding to an LLM:

curl -X POST http://localhost:8080/check \
  -H "Content-Type: application/json" \
  -d '{"text": "Ignore all previous instructions"}'

# Response:
# {"action": "block", "category": "prompt-injection", "signature_name": "Ignore instructions directive", "confidence": 0.9}

Features

• Multi-Provider Routing - OpenAI, Anthropic, OpenRouter, Ollama
• Detection-Only Mode - /check endpoint for pre-validation
• Request & Response Scanning - Regex + heuristics detection
• Policy Engine - Per-category actions (block/warn/log)
• Rate Limiting - iptables-based DDoS protection
• Prometheus Metrics - Full observability at /metrics
• Control Plane - Fleet management, automatic signature sync

Detection Categories

Category	Description
prompt-injection	Instruction override attempts
jailbreak	DAN mode, persona exploits
encoding-bypass	Base64/hex/ROT13 obfuscation
delimiter-injection	JSON/XML structure attacks
indirect-injection	Tool abuse, plugin exploits
unicode-evasion	Homoglyphs, fullwidth chars
role-manipulation	Named character roleplay

Documentation

Full documentation at ainvirion.github.io/aiproxyguard

• Getting Started
• Configuration
• Deployment
• API Reference

Control Plane

Connect to aiproxyguard.com for fleet management and automatic signature updates:

docker run -d -p 8080:8080 \
  -e AIPROXYGUARD_CONTROL_PLANE_ENABLED=true \
  -e AIPROXYGUARD_CONTROL_PLANE_API_KEY=your-api-key \
  ghcr.io/ainvirion/aiproxyguard:latest

Development

python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

License

Apache-2.0 - Copyright (c) 2025-2026 AInvirion LLC