Ollama Pentest AI - Burp Suite Extension

Enhanced fork of DeepSeek Pentest AI with local Ollama support

Original Author: Hernán Rodríguez

Original Project: DeepSeek-Pentest-AI

LinkedIn: https://www.linkedin.com/in/hernanrodriguez-/

---

⚠️ Disclaimer

This tool is intended for EDUCATIONAL and AUTHORIZED security testing purposes ONLY.

• Do NOT use against systems without explicit written permission
• The authors are NOT responsible for misuse of this extension
• Use ONLY on systems you own or have authorization to test
• Recommended for local testing against vulnerable apps like OWASP Juice Shop or DVWA
• USE AT YOUR OWN RISK - NO WARRANTY PROVIDED

---

🎯 What's New in Ollama Version

This fork replaces the cloud-based DeepSeek API with local Ollama models , giving you:

• ✅ Privacy : All payload generation happens locally
• ✅ No API costs : Use uncensored models for offensive security testing
• ✅ Full control : Choose from multiple Ollama models
• ✅ Educational focus : Perfect for learning penetration testing techniques

Version Comparison

Feature
v1 (Basic)
v2 (Enhanced)
Local Ollama Support
✅
✅
Context-Aware Payloads
❌
✅
Risk Scoring (20-95%)
❌
✅
Color-Coded Risk Badges
❌
✅
Multi-Model Support
❌
✅
Safe Mode (No Destructive)
❌
✅
Session Save/Load
❌
✅
AI Summary Generation
❌
✅
Temperature Control
❌
✅

Recommended Version : v2 for production use

---

🚀 Key Features

Core Capabilities

• AI-Powered Payload Generation : Uses local Ollama models for intelligent payload creation
• Multiple Attack Types : SQL Injection, XSS, Command Injection, Path Traversal, LFI, SSRF, RCE, SSTI, XXE, NoSQL, GraphQL, Open Redirect, CRLF, CORS, Host Header Injection
• Smart Parameter Detection : Automatically finds parameters in GET, POST, JSON, XML, multipart, and headers
• Intelligent Fuzzing : Baseline comparison, differential analysis, confidence scoring
• Real-Time Metrics : Live vulnerability tracking with visual charts

Enhanced v2 Features

• 🎯 Context-Aware Payloads : Extracts actual parameter values for targeted payload generation
• 🎨 Risk Badges : Color-coded severity indicators (Critical/High/Medium/Low)
• 🛡️ Safe Mode : Filters destructive SQL payloads (DROP, DELETE, UPDATE, etc.) with confirmation dialog
• 🧠 Multiple Models : Choose from mistral, llama2, phi3, neural-chat, gdisney/mistral-uncensored
• 🌡️ Fine-Tuning : Adjust temperature and max tokens for optimal results
• 💾 Session Persistence : Save/load entire testing sessions as JSON
• 📊 AI Summaries : Generate executive reports with remediation recommendations
• 📈 Confidence Scoring : 20-95% confidence ratings based on evidence strength

---

📋 Prerequisites

Required

• Burp Suite Pro (Community edition has limited functionality)
• Ollama installed locally (Download)
• Jython standalone JAR configured in Burp Extender
• Python 2.7 environment (for Jython compatibility)

Recommended Ollama Model

bash

ollama pull gdisney/mistral-uncensored

Note: This is an uncensored model suitable for offensive security research. Use responsibly.

---

🔧 Installation

1. Install Ollama

bash

# macOS/Linux
curl https://ollama.ai/install.sh |sh

# Windows
# Download from https://ollama.ai/download

2. Pull a Model

bash

ollama pull gdisney/mistral-uncensored
# or
ollama pull mistral
ollama pull llama2

3. Configure Burp Suite

1. Load Jython:

   • Extender → Options → Python Environment
   • Set location of Jython standalone JAR

2. Add Extension:

   • Extender → Extensions → Add
   • Extension type: Python
   • Select: Ollama-Pentest-AI-localhost-v2.py(recommended)
   • Check Burp Extender output for: Plugin initialized

4. Configure the Extension

Basic Setup:

• API Key: Enter any value (e.g., "localhost" - required by UI but not used)
• Model: Select gdisney/mistral-uncensored or your preferred model
• Attack Type: Choose vulnerability type or "CUSTOM PROMPT"
• Payload Count: 5-10 recommended
• Delay: 100ms (adjust based on target)

Advanced Settings (v2 only):

• Temperature: 0.4(lower =more focused, higher =more creative)
• Max Tokens: 800(longer for complex payloads)
• ✅ Safe Mode: ENABLED BY DEFAULT - prevents destructive SQL
• ✅ Smart Fuzzing: Enable for adaptive payload generation
• ✅ Capture all traffic: Monitor all Burp requests

---

📖 Usage Guide

Basic Workflow

1. Capture a Request

• Use Burp Proxy to intercept a request
• Right-click → "Ollama: Analyze Request"
• Or manually paste a request into the extension tab

2. Generate Payloads

1. Select Attack Type (e.g., "SQL Injection")
2. Click "Analyze & Generate"
3. Review generated payloads in"AI Analysis" tab
4. Parameters automatically detected and displayed

3. Start Pentesting

1. Click "Start Pentesting"
2. Watch real-time results in"Pentest Live" tab
3. Color-coded vulnerabilities appear (v2: with risk badges)
4. Check "Results" tab for detailed findings

4. Review & Export

- View metrics in"Metrics" tab
- Generate AI summary (v2): "AI Summary" → "Generate Summary"
- Export findings: "Export to CSV"
  - Only vulnerabilities
  - Full history
  - Evidence-only

Custom Prompts

For specialized testing, use CUSTOM PROMPT mode:

Attack Type: CUSTOM PROMPT
Custom Prompt: "Generate advanced SQL injection payloads for boolean-based blind attacks that bypass ModSecurity WAF. Return ONLY payload values, one per line, no explanations."

Important: Always end custom prompts with: "Return ONLY payload values, one per line, no explanations."

---

🔒 Safe Mode Feature (v2)

What is Safe Mode?

Safe Mode filters out destructive SQL payloads that could cause permanent data loss:

Blocked Operations:

• DROP(tables/databases)
• DELETE(row deletion)
• UPDATE(data modification)
• INSERT(data injection)
• TRUNCATE(table clearing)
• ALTER(schema changes)
• CREATE, RENAME, GRANT, REVOKE

Using Safe Mode

Default: ✅ ENABLED (Recommended)

To Disable (not recommended):

1. Uncheck "Safe Mode (no destructive)"
2. Confirmation dialog appears with warning
3. Click "Yes" to confirm (use extreme caution)

Best Practice: Keep Safe Mode enabled unless you have explicit authorization and understand the risks.

---

📊 Understanding Results

Risk Ratings (v2)

Color	Risk Level	Severity	Example
🔴 Dark Red	Critical	RCE, SSRF, XXE	Remote code execution
🟠 Orange	High	SQL Injection, XSS	Data exfiltration possible
🟡 Yellow	Medium	Open Redirect, CORS	Limited impact
🟢 Green	Low	Information Disclosure	Minimal risk

Confidence Scores (v2)

• 20-50%: Low confidence, possible false positive
• 51-74%: Medium confidence, worth investigating
• 75-89%: High confidence, likely vulnerable
• 90-95%: Very high confidence, exploit confirmed

Evidence Indicators

Findings marked with "Evidence:" show actual proof:

SQL Injection - Database error detected [high|85%]| Evidence: "mysql_fetch_array() expects"

---

🎛️ Advanced Configuration

Model Selection

Recommended Models:

1. gdisney/mistral-uncensored (Best for offensive security)

   • Uncensored, generates aggressive payloads
   • Great for educational testing

2. mistral (Balanced)

   • Good quality, moderate censorship
   • Suitable for general testing

3. llama2 (Conservative)

   • More censored, safer outputs
   • Good for compliance testing

Temperature Settings

0.1-0.3: Focused, deterministic payloads
0.4-0.6: Balanced creativity (recommended)
0.7-0.9: Highly creative, diverse payloads
1.0+: Experimental, unpredictable

Performance Tuning

Request Delay: 
  - 0ms: Maximum speed (may trigger WAF)
  - 100ms: Balanced (recommended)
  - 500ms+: Stealth mode

Payload Count:
  - 5: Quick scan
  - 10: Standard test
  - 20+: Comprehensive audit

---

🔍 Comparison: v1 vs v2

Choose v1 If:

• ✅ You need basic functionality
• ✅ You're new to Burp extensions
• ✅ You prefer simplicity

Choose v2 If:

• ✅ You need production-ready features
• ✅ You want risk scoring and context-aware payloads
• ✅ You need session persistence
• ✅ You want AI-generated reports
• ✅ You need Safe Mode protection

Recommendation: Use v2 for all serious testing scenarios.

---

📁 Project Structure

Ollama-Pentest-AI/
├── Ollama-Pentest-AI-localhost-v1.py  # Basic version
├── Ollama-Pentest-AI-localhost-v2.py  # Enhanced version (recommended)
├── DeepSeek-Pentest-AI.py             # Original DeepSeek version
├── README-v1.md                        # v1 documentation
├── README-v2.md                        # v2 documentation
└── README.md                           # This file

---

🐛 Troubleshooting

Common Issues

"API test failed"

bash

# Check Ollama is running
ollama list

# Start Ollama service
ollama serve

# Test with curl
curl http://localhost:11434/v1/chat/completions

"No parameters detected"

• Ensure request starts with HTTP method (GET/POST/PUT/DELETE)
• Check Content-Type header is present
• Try capturing request through Burp Proxy

"Generate payloads first"

• Click "Analyze & Generate" before "Start Pentesting"
• Wait for payload generation to complete
• Check "AI Analysis" tab for generated payloads

Slow payload generation

python

# Reduce max_tokens in UI
Max Tokens:400(instead of 800)

# Or use faster model
Model: mistral (instead of llama2)

Extension not loading

• Verify Jython is configured in Burp
• Check Burp Extender "Output" tab for errors
• Ensure Python file is not corrupted

---

🤝 Contributing

This is an educational fork. Contributions welcome:

1. Fork the repository
2. Create a feature branch
3. Test thoroughly on vulnerable apps (DVWA, Juice Shop)
4. Submit pull request with detailed description

---

📜 License

This project inherits the license from the original DeepSeek-Pentest-AI repository.

---

🙏 Credits

Original Author : Hernán Rodríguez

• GitHub: @HernanRodriguez1
• Original Project: DeepSeek-Pentest-AI
• LinkedIn: hernanrodriguez-

Ollama Adaptation : Educational fork for local AI-powered pentesting

---

📚 Further Reading

• Burp Suite Extensions Documentation
• Ollama Documentation
• OWASP Testing Guide
• Jython for Burp

---

⚖️ Legal Notice

EDUCATIONAL USE ONLY

This tool is designed for:

• ✅ Authorized penetration testing
• ✅ Security research in controlled environments
• ✅ Educational demonstrations
• ✅ Bug bounty programs with permission

NEVER USE FOR :

• ❌ Unauthorized access attempts
• ❌ Malicious attacks
• ❌ Testing without permission
• ❌ Any illegal activity

By using this tool, you agree :

• You have explicit authorization to test target systems
• You understand the legal implications
• You accept full responsibility for your actions
• You will use Safe Mode in production environments

Runs in Burp Pro or Community

---

Happy (Legal) Hacking! 🎯

Remember: With great power comes great responsibility. Use this tool ethically.