OllaDeck is a purple technology stack for Generative AI (text modality) cybersecurity. It provides a comprehensive set of tools for both blue team and red team operations in the context of text-based generative AI.
- First release of OllaDeck is now available!
- Setup verified with AMD64 architecture, Docker for Desktop on Windows, WSL2 with Ubuntu, and local storage.
- Overview
- Key Components
- Code of Conduct
- Blue Team
- Red Team
- Installation
- Usage
- News
- Planned Items
- Contributing
- License
OllaDeck is a purple technology stack for Generative AI (text modality) security. It combines LLM platforms, various database types, and specialized tools for both offensive and defensive cybersecurity operations.
- LLM platforms for loading, managing, and performing LLM operations (e.g., chat)
- Vector, graph, and traditional databases for data operations
- Red team instance with tools for Generative AI (text modality)
- Blue team instance with Generative AI-based defense tools
- Sample endpoints for simulated attacks and defenses
Please note that this project is released with a Contributor Code of Conduct. By participating in this project and/or cloning the project, you agree to abide by its terms.
The blue team in the context of text-based generative AI focuses on defending and securing AI systems. This includes activities such as:
- Monitoring AI model outputs for potential vulnerabilities
- Implementing safeguards against prompt injection and other AI-specific attacks
- Ensuring compliance with ethical AI guidelines and regulations
- Analyzing AI-generated content for potential security risks
Currently, OllaDeck has the following Blue Team tools.
- AutoGluon: Automates machine learning tasks
- AutoViz: Automated visualization library
- python-dateutil: Powerful extensions to the standard datetime module
- lxml: Library for processing XML and HTML
- PyPDF2: PDF file manipulation
- docx: Python library for creating and updating Microsoft Word (.docx) files
- openpyxl: Python library to read/write Excel files
- Tesseract: Optical Character Recognition engine
- TextBlob: Simplified text processing
- Gensim: Topic modeling, document indexing, and similarity retrieval
- Compliance Masonry: Tool for building compliance documentation
- CybOX: Structured language for cyber observables
- STIX: Structured language for cyber threat intelligence
The red team in text-based generative AI focuses on identifying vulnerabilities and potential attack vectors in AI systems. Activities include:
- Crafting adversarial prompts to manipulate AI outputs
- Testing AI model robustness against various attack techniques
- Exploring potential data leakage or privacy concerns in AI responses
- Simulating real-world attack scenarios on AI systems
Currently, OllaDeck has the following Blue Team tools.
- pyRIT: Red team automation toolkit
- TextAttack: Framework for adversarial attacks on NLP models
- Adversarial Robustness Toolbox: Library for machine learning security
- Garak: LLM vulnerability scanner
For detailed installation instructions, please refer to our Installation Guide.
- Add more JupyterNotebook examples
- Add more datasets
- Enhanced documentation and user guides
- Add more tools for both blue and red team operations
We welcome contributions to OllaDeck! Please see our Contributing Guide for more information on how to get started.