Rebuff: LLM Prompt Injection Detector
Rebuff is an LLM Prompt Injection Detector developed to protect AI applications from prompt injection (PI) attacks through a multi-layered defense. It is designed as a prototype and offers several key features:
Key Features:
- Prompt Injection Detection: Identifies potential prompt injection attacks using heuristics and LLM-based detection.
- Canary Word Leak Detection: Adds canary tokens to prompts, enabling detection of leakage and enhancing security.
- Self-hosting: Provides options for users to self-host the detector, offering control and flexibility in deployment.
- Integration Support: Compatible with cloud services like OpenAI and Pinecone for seamless integration.
Benefits:
- Multi-layered Defense: Four layers of defense to ensure enhanced security against prompt injection attacks.
- Open Source: Contributions from the community are welcome, fostering collaboration and improvement.
- User-defined Strategies: Customize detection strategies to meet specific needs.
Highlights:
- Written in TypeScript and Python for compatibility across platforms.
- Supports flexible deployment options, allowing users to tailor their experience based on requirements.
- Encourages community involvement through contributions and feedback.