sandstrike 1.0.3

A comprehensive Python library and CLI tool to perform LLM Red Teaming with Avenlis SandStrike.

---

Avenlis SandStrike

Avenlis SandStrike is a comprehensive Python library and CLI tool for AI red team testing and LLM security assessment. Features include comprehensive adversarial prompt library, automated vulnerability detection, persistent session management, and a modern web interface.

Features

• 🔒 Advanced Red Teaming: Comprehensive library of adversarial prompts covering multiple attack techniques
• 🔧 Encoding Support: Coming soon (currently not available)
• 📁 Smart Collections: Organize and manage adversarial prompts with intelligent categorization
• 🎯 Local LLM Testing: Full Ollama integration for testing local models
• 🌐 Platform Integration: Fetch prompts from Avenlis platform
• 📊 Session Management: Track and analyze test results over time
• 🖥️ Modern Web Interface: React-based UI with real-time updates, no authentication required
• ⚡ CLI Tools: Powerful command-line interface for automation and scripting
• 🔍 Intelligent Prompt Analysis: Automatic classification and OWASP mapping
• 💾 Persistent Data: Save test results and session data locally (encoding persistence coming soon)
• 📈 SandStrike Copilot Grader: AI-powered response evaluation using production LLM (Pro users only)

Installation

From PyPI

pip install sandstrike

From Source

git clone https://github.com/StateraSolutions/avenlis-sandstrike.git
cd avenlis-sandstrike
pip install -e .

Quick Start

1. Install the package:

pip install -e .

2. Start the web interface:

sandstrike ui start

3. Open your browser and navigate to http://localhost:3000

Usage

Web Interface

Start the web interface:

sandstrike ui start                    # Start on default port
sandstrike ui start --port 3000        # Start on custom port

The web interface provides:

• Dashboard: Real-time overview of security metrics
• Sessions: View and manage testing sessions with detailed analytics
• Collections: Organize prompts into logical groups
• Prompts: Browse and manage adversarial prompts
• OWASP LLM & MITRE ATLAS: Browse industry-standard vulnerability frameworks

CLI Commands

# Authentication
sandstrike auth verify

# Prompt Management
sandstrike prompts list
sandstrike prompts create --id ID --technique TECH --category CAT --text TEXT --source local
sandstrike prompts view PROMPT_ID
sandstrike prompts delete PROMPT_ID

# Collections Management
sandstrike collections list
sandstrike collections create --name "Custom Tests" --description "My test suite" --source local
sandstrike collections add-prompt --collection-id ID --collection-source local --prompt-ids prompt1,prompt2 --prompt-source file
sandstrike collections remove-prompt --collection-id ID --collection-source local --prompt-ids prompt1
sandstrike collections delete COLLECTION_ID

# Target Management
sandstrike targets list
sandstrike targets create --id TARGET_ID --name "My Target" --ip "http://localhost:11434" --source local
sandstrike targets view --target-ids TARGET_ID --source local
sandstrike targets delete --target-ids TARGET_ID --source local

# Session Management
sandstrike sessions list
sandstrike sessions show SESSION_ID --detailed

# Reports (Pro users only)
sandstrike reports overview --session-id session_1,session_2
sandstrike reports detailed --source file
sandstrike reports executive --session-id session_1,session_2
sandstrike reports all --session-id session_1,session_2

# Dynamic Variables
sandstrike variables list --source all
sandstrike variables set --category application --name banking_app --value "SecureBank" --source local
sandstrike variables update --category application --name banking_app --value "NewName" --source local

# Database Management
sandstrike database status
sandstrike database local-wipe

# Get help
sandstrike --help
sandstrike [command] --help

Python API

from avenlis.redteam import AvenlisRedteam, RedteamSession
from avenlis.config import AvenlisConfig
from avenlis.main_storage import AvenlisStorage

# Configuration
config = AvenlisConfig()

# Red Team Testing
redteam = AvenlisRedteam()

# Run security tests against Ollama
results = redteam.run_attacks(
    target="ollama://llama3.2:1b@localhost:11434",
    prompt_ids=["prompt_001", "prompt_002"]
)

print(f"Attack success rate: {results['summary']['success_rate']:.1f}%")

# Session management
session = RedteamSession.create("Security Audit", "ollama://llama3.2:1b@localhost:11434")

# Storage and data management
storage = AvenlisStorage(config)
sessions = storage.get_combined_sessions()
prompts = storage.get_combined_prompts()

Configuration

Create a .env file in the project root directory:

# Avenlis Platform Configuration
AVENLIS_API_URL=https://api.avenlis.com
AVENLIS_API_KEY=your_api_key_here

# Local Testing Configuration
OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama2

# Web Interface Configuration
WEB_PORT=8080
WEB_HOST=0.0.0

Configuration is automatically loaded from environment variables, .env file, or default values.

Troubleshooting

Web Interface Not Starting

sandstrike ui start --port 3000  # Try different port

Import Errors

pip install -e . --force-reinstall

Database Issues

sandstrike database status
sandstrike database local-wipe

Getting Help

• User Guide: See USER_GUIDE.md for detailed usage instructions
• CLI Reference: Use sandstrike --help for command information
• CLI Cheat Sheet: See CLI_CHEATSHEET.md for quick command reference
• Issues: Report bugs and feature requests on GitHub
• Discord: Join our community for support and discussions

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

---

Avenlis SandStrike - Empowering AI Security Testing

Website • User Guide • CLI Reference • Install Guide