🚀 CodeGlia – Intelligent Vulnerability Scanner

CodeGlia is an AI-augmented static analysis tool that integrates Semgrep, Bandit, and LLM-based scanning (Google Gemini, OpenAI GPT-4, or Ollama) to detect and explain vulnerabilities across multiple programming languages.

---

🧠 Overview

CodeGlia provides real-time vulnerability scanning through an interactive Flask-based dashboard.
It allows developers to upload entire projects (ZIP or folder), scans the source code for potential security issues, and generates AI-enriched reports with performance comparison metrics.

---

🧩 Features

• 🔍 Multi-language static analysis (Python, Java, C, C#, PHP, JavaScript)
• 🤖 AI-powered vulnerability explanation and severity labeling
• 🚨 NEW: LLM-based vulnerability detection with multiple backend support
• 📊 Performance comparison between static and LLM analysis
• 📈 Trust Score computation and visual trend tracking
• 🌐 Flask-based web dashboard with live scan progress
• 🧱 Dockerized for fast deployment and consistency
• ⚙️ Supports local Semgrep rule packs for offline scanning

---

🤖 LLM Scanner Features

Supported Backends

• Google Gemini (default) - Fast and efficient with gemini-2.0-flash-exp
• OpenAI GPT-4 - High-quality analysis (requires API key)
• Ollama - Local models (CodeLlama, Mistral) for offline scanning

Capabilities

• Scans multiple file types: .py, .js, .java, .php, .c, .cpp, .rb, etc.
• Detects: SQL injection, command injection, XSS, path traversal, hardcoded secrets, weak crypto, and more
• Maps findings to CWE/CVE using existing infrastructure
• Implements caching to avoid re-analyzing identical code
• Provides detailed explanations and secure fix recommendations

Performance Metrics

The scanner tracks and compares:

• Individual timing for each scanner (Bandit, Semgrep, LLM)
• Throughput metrics (lines/second)
• Total static vs LLM analysis time
• Time difference and speedup/slowdown factors

---

🖥️ Tech Stack

Layer	Technology
Frontend	HTML, CSS (custom dark theme), JS
Backend	Python (Flask)
AI Integration	Google Gemini API, OpenAI (optional), Ollama (optional)
Security Analysis	Semgrep, Bandit, LLM Scanner
Packaging & Deployment	Docker
Data Handling	JSON, Pandas (for parsing reports)

---

🧪 How to Run Locally

Option 1 – Using Python

git clone https://github.com/Shreyas-135/CODEGLIA.git
cd CODEGLIA
pip install -r requirements.txt
flask run

Option 2 – Using Docker

docker build -t codeglia .
docker run -p 5050:5050 --env-file .env codeglia

Then open your browser at:
👉 http://localhost:5050

---

⚙️ Environment Variables

Create a file named .env in the project root with the following content:

# Flask configuration
FLASK_APP=app.py
FLASK_RUN_HOST=0.0.0.0

# AI Configuration
GEMINI_API_KEY=your_api_key_here

# LLM Scanner Configuration (optional)
LLM_ENABLED=false                    # Set to true to enable LLM scanning
LLM_BACKEND=gemini                   # Options: gemini, openai, ollama
LLM_MODEL=gemini-2.0-flash-exp       # Model name (optional, uses defaults)

# Optional: OpenAI Configuration
# OPENAI_API_KEY=your_openai_key_here

# Optional: Ollama Configuration (for local models)
# OLLAMA_BASE_URL=http://localhost:11434

---

🚀 CLI Usage

Basic Scan (Static Analysis Only - Default)

python run_scan.py datasets/

Enable LLM Scanning

python run_scan.py datasets/ --enable-llm

Run with Specific LLM Backend

python run_scan.py datasets/ --enable-llm --llm-backend gemini --llm-model gemini-2.0-flash-exp

Compare All Scanners

python run_scan.py datasets/ --compare

Static Analysis Only (Bandit + Semgrep)

python run_scan.py datasets/ --static-only

LLM Scanner Only (for testing)

python run_scan.py datasets/ --llm-only --llm-backend gemini

CLI Options

• --enable-llm - Enable LLM scanning
• --llm-backend [gemini|openai|ollama] - Choose LLM backend
• --llm-model <model-name> - Specify model
• --static-only - Run only static analyzers (default for backward compatibility)
• --llm-only - Run only LLM scanner (for testing)
• --compare - Run all scanners and generate comparison report

---

📊 Output

After scanning, reports are generated in the output/ folder:

• scan_report.json – Raw vulnerability data
• scan_report.html – Human-readable HTML report with performance comparison
• performance.json – Detailed metrics on scan time, throughput, and comparisons

Example Performance Output

===== Starting CodeGlia Workflow =====
📁 Using target dataset directory: datasets
🤖 LLM scanning enabled (backend: gemini, model: gemini-2.0-flash-exp)

🚀 Running Bandit Scan (Python)...
✅ Bandit completed in 2.5s (5 vulnerabilities found)

🚀 Running Semgrep Scan (Multi-language)...
✅ Semgrep completed in 8.3s (12 vulnerabilities found)

🚀 Running LLM Scan (gemini/gemini-2.0-flash-exp)...
📊 Processing files in datasets...
✅ LLM completed in 45.2s (8 vulnerabilities found)

📊 Performance Comparison:
   Static Analysis Total: 10.8s
   LLM Analysis Total: 45.2s
   LLM is 4.2x slower (+34.4s)
   
   Throughput:
   - Bandit: 494 lines/sec
   - Semgrep: 149 lines/sec
   - LLM: 27 lines/sec

✅ Total scan time: 56.0s

---

🧱 Folder Structure

CodeGlia/
│
├── app.py                # Flask backend
├── run_scan.py           # Handles Semgrep + Bandit + LLM scanning
├── parse_results.py      # AI-driven report parsing
├── llm_scanner.py        # LLM-based vulnerability scanner module
├── requirements.txt      # Dependencies
├── Dockerfile            # Docker configuration
│
├── static/               # CSS, JS, icons
├── templates/            # HTML templates (Flask views)
└── output/               # Generated reports

---

🌟 LLM Model Selection Best Practices

For Speed

• Gemini 2.0 Flash - Fast, cost-effective, good accuracy
• Ollama with CodeLlama 7B - Local, fast, no API costs

For Accuracy

• GPT-4 - Best accuracy, slower, higher cost
• Gemini Pro - Good balance of speed and accuracy

For Offline/Privacy

• Ollama with Mistral - Run locally, no internet required
• Ollama with CodeLlama - Optimized for code analysis

---

🔒 Security Considerations

• LLM scanner is disabled by default to maintain backward compatibility
• API keys are never logged or stored in scan results
• Cache files use content hashes, not source code
• All scanners can run independently if one fails

---

🌍 Deployment

CodeGlia can be deployed easily on:

• Render
• Railway
• Oracle Cloud (OCI)
• Azure App Service
• Amazon Web Services

Each instance supports .env for secure API key handling.