The world's first comprehensive AI defense platform that proves vulnerabilities with live attacks and fixes them automatically.

AEGIS (Advanced Electronic Guard & Intelligence System) is powered by Google Gemini 3 Pro to perform deep static and dynamic analysis of neural networks across multiple frameworks (PyTorch, TensorFlow, Keras, Go).

🎥 Watch Demo Video

• Red Team Mode: Generates working exploit code (FGSM attacks, DoS vectors)
• Blue Team Mode: Generates remediated code with security patches
• Code Diff Viewer: Side-by-side comparison showing exact fixes
• Only tool that generates both the attack AND the fix

• Real-time FGSM attack visualization
• Watch model confidence degrade (98% → 2%)
• Animated attack success rate (87%)
• Visual proof of vulnerabilities

• AI improving AI through reinforcement learning
• Live episode streaming (Episode 1: +10 reward, Episode 2: +15...)
• Security improvements: 50/100 → 85/100 in 5 iterations
• Impossible before Gemini 3 Pro

• Multi-Language CI/CD Agents: Python, Node.js, Go
• EU AI Act Compliance Certificates: Downloadable PDF reports
• Threat Intelligence Feed: Real-time security statistics
• System Architecture Visualization: Interactive flowcharts

• Vulnerability Detection: Hardcoded shapes, adversarial susceptibility, missing regularization
• Architecture Analysis: Layer-by-layer breakdown, bottleneck identification
• Performance Metrics: Parameters, FLOPs, memory usage, efficiency
• Explainability: Decision process, feature importance, interactive Q&A

# 1. Fork the repository on GitHub (click Fork button)

# 2. Clone your fork
git clone https://github.com/YOUR_USERNAME/aegis-ai-defence.git
cd aegis-ai-defence

# 3. Set your Gemini API key
echo "GEMINI_API_KEY=your_api_key_here" > .env

# 4. Run with Docker
docker-compose up -d

# 5. Access AEGIS
open http://localhost:3000

# Install dependencies
npm install

# Set API key
export GEMINI_API_KEY=your_api_key_here

# Run development server
npm run dev

# Access AEGIS
open http://localhost:3000

# Install dependencies
pip install google-generativeai

# Run audit
python cli/python/aegis_audit.py \
  --file your_model.py \
  --api-key YOUR_GEMINI_KEY \
  --threshold 80

aegis-ai-defence/
├── components/          # React UI components
│   ├── AnalysisPanel.tsx       # Main dashboard
│   ├── AttackSimulator.tsx     # Live attack visualization
│   ├── ActiveDefense.tsx       # Red/Blue team modes
│   ├── ComplianceCertificate.tsx
│   └── ...
├── services/
│   └── geminiService.ts        # Gemini 3 Pro integration
├── cli/
│   ├── python/                 # Python CLI auditor
│   ├── nodejs/                 # Node.js CLI auditor
│   └── go/                     # Go CLI auditor
├── docs/                       # Documentation
├── docker-compose.yml          # Docker setup
├── Dockerfile
└── README.md

Audit diagnostic models for bias and explainability before deployment. Generate FDA/CE compliance documentation.

Validate credit scoring models for fairness. Detect adversarial manipulation of fraud detection systems.

Security audit for safety-critical AI. Prevent adversarial attacks on perception systems.

Ensure recruitment AI is unbiased. Explain hiring decisions for legal compliance.

Verify fairness across user groups. Detect manipulation attempts.

name: AI Security Gate

on: [push, pull_request]

jobs:
  aegis-audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
      
      - name: Run AEGIS Audit
        run: |
          python cli/python/aegis_audit.py \
            --file ./models/net.py \
            --api-key ${{ secrets.GEMINI_KEY }} \
            --threshold 80

aegis-audit:
  script:
    - python cli/python/aegis_audit.py --file model.py --api-key $GEMINI_KEY --threshold 80
  only:
    - main

Input: SimpleCNN (MNIST classifier)

Output:

• Security Score: 50/100
• Vulnerabilities: 4 (1 Critical, 1 High, 2 Medium)
  • ❌ CRITICAL: Hardcoded input shape (DoS vulnerability)
  • ⚠️ HIGH: Adversarial susceptibility (87%)
  • ⚠️ MEDIUM: Missing regularization
  • ⚠️ MEDIUM: Parameter bottleneck (94% in fc1 layer)

Generated:

• ✅ exploit_fgsm.py - Working FGSM attack script
• ✅ secure_model.py - Fixed model with patches
• ✅ compliance_certificate.pdf - EU AI Act report

• Frontend: React 19, TypeScript, Tailwind CSS
• AI Engine: Google Gemini 3 Pro
• Visualization: Recharts
• Video Generation: Google Veo 3.1
• Deployment: Docker, Vite
• Testing: Vitest, React Testing Library

1. Code Upload: Paste or upload neural network code
2. Gemini Analysis: Streaming analysis with structured JSON schemas
3. Vulnerability Detection: Identifies security flaws with severity scoring
4. Active Defense: Generates exploit code + remediated code
5. RL Optimization: Iteratively improves model security
6. Compliance: Generates EU AI Act certificates

python cli/python/aegis_audit.py --file model.py --api-key KEY --threshold 80

node cli/nodejs/aegis-audit.js --file model.js --api-key KEY --threshold 80

go run cli/go/aegis-audit.go --file model.go --api-key KEY --threshold 80

• Installation Guide
• User Guide
• API Reference
• CI/CD Integration
• Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

# Fork the repository
# Create a feature branch
git checkout -b feature/amazing-feature

# Commit your changes
git commit -m 'Add amazing feature'

# Push to the branch
git push origin feature/amazing-feature

# Open a Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.

• Demo Video: https://youtu.be/QBqkJdfmxhk
• Documentation: docs/
• Issues: GitHub Issues

• Author: Pralay Ghosh
• Email: [email protected]

• Google DeepMind for Gemini 3 Pro API
• Google Veo for video generation capabilities
• Open Source Community for amazing tools and libraries

AEGIS is a security auditing tool. The exploit code generated is for educational and security testing purposes only. Always obtain proper authorization before testing systems you don't own.