Prompt Fence SDK
Secure your specific LLM prompts with cryptographic boundaries.
The Prompt Fence SDK provides an easy-to-use API for implementing cryptographic trust boundaries in LLM applications. It allows you to mathematically prove which parts of a prompt came from your trusted system and which parts came from untrusted users.
[!NOTE] This is an unofficial implementation of the concept described in the paper Prompt Fence: A Cryptographic Approach to Establishing Security Boundaries in Large Language Model Prompts.
Why Prompt Fence?
LLMs cannot inherently distinguish between "instructions" and "data". This vulnerability leads to Prompt Injection, where user input hijacks the model's behavior.
Prompt Fence solves this by:
- Signing: Your application cryptographically signs trusted instructions using Ed25519.
- Fencing: Trusted and untrusted content is wrapped in XML "fences" with digital signatures.
- Verifying: A security gateway (or the LLM itself) validates these signatures before processing.
Core Features
- 🛡️ Cryptographic Integrity: Uses Ed25519 signatures to prevent tampering.
- ⚡ High Performance: Core cryptography is implemented in Rust.
- 🔒 Zero-Trust Architecture: Treat all inputs as untrusted until validated.
- 📝 Flexible Builders: Easy-to-use Python API for constructing complex prompts.
Get Started
Ready to secure your prompts?