🛡️ AI Security Assistant

Enterprise-grade AI-powered security assistant with agentic capabilities, comprehensive audit logging, and advanced observability. https://www.loom.com/share/444a8ace9d26404b8a2614528757a372?sid=1141c83c-6bc4-4346-b8e2-a976949a365f

🎯 Overview

An intelligent security assistant that combines Generative AI (LLM-powered conversations) with Agentic AI (autonomous tool selection) to help security teams with incident response, policy guidance, and threat intelligence.

✨ Key Features

• 🤖 Agentic AI: Autonomous decision-making and tool selection using LangGraph
• 🔍 Smart Tools: Policy search, log analysis, and real-time web search
• 🛡️ Enterprise Security: Role-based access control, DLP, prompt injection defense
• 📊 Observability: LangSmith integration with full tracing and transparency
• 💬 Memory: Multi-turn conversations with persistent context
• 🎨 Rich UI: Markdown rendering with security transparency explanations

🚀 Quick Setup

# 1. Clone the repository
git clone https://github.com/Shubham-Saboo/AI-security-assistant.git
cd AI-security-assistant

# 2. Setup and start
./start.sh setup && ./start.sh

Access the application: http://localhost:3000

📋 Available Commands

./start.sh setup      # Initial environment setup
./start.sh            # Start both backend and frontend
./start.sh backend    # Start backend only
./start.sh frontend   # Start frontend only  
./start.sh status     # Check service status
./start.sh stop       # Stop all services
./start.sh restart    # Restart everything
./start.sh --help     # Show all options

🔑 Configuration

Add your API keys to backend/.env:

# Required
OPENAI_API_KEY=your_openai_api_key_here

# Optional (for enhanced features)
LANGSMITH_API_KEY=your_langsmith_api_key_here    # Enhanced observability
TAVILY_API_KEY=your_tavily_api_key_here          # Web search capabilities

🧪 Demo Scenarios

Security Role

• "How should I handle a phishing email?"
• "Show me recent failed login attempts"
• "What are the latest CVE vulnerabilities?" (requires web search)

Sales Role

• "What are our password requirements?"
• "Show me security logs" (limited access)

🔒 Security Features

• Role-Based Access Control: Security vs Sales permissions
• Data Loss Prevention: Automatic masking of sensitive data
• Prompt Injection Defense: Malicious input detection
• Audit Logging: Complete action tracking with trace correlation
• Security Transparency: "Why did I answer this way?" explanations

📊 Architecture

Built with:

• Backend: FastAPI + LangChain + LangGraph
• Frontend: React + TypeScript
• AI: OpenAI GPT-4o-mini with agentic capabilities
• Data: ChromaDB (vector search) + CSV logs + Markdown policies
• Observability: LangSmith integration

For detailed architecture and design notes, see ARCHITECTURE.md.

📚 Documentation

• DEMO_GUIDE.md: Complete testing scenarios and examples
• ARCHITECTURE.md: Technical design and implementation details