Rebuff
Open-source four-layer prompt injection detector for LLM apps.
About
Rebuff is an open-source prompt injection detector from Protect AI that stacks four defensive layers: heuristic filtering, an LLM-based classifier, a vector database of known attack embeddings for similarity matching, and canary tokens to catch prompt leakage. It ships as Python and TypeScript SDKs.
It is explicitly labeled a prototype by its authors and does not claim complete protection against prompt injection. It pairs naturally with broader guardrail toolkits like LLM Guard for input and output sanitization.
What it can do
- Heuristic plus LLM-based injection detection
- Vector similarity against known attacks
- Canary token leakage detection
- Python and TypeScript SDKs
Setup
pip install rebuff, then wire detect_injection() into your prompt pipeline with a Pinecone index.
Prerequisites
Python or TypeScript project, OpenAI API key, and a vector database for attack signatures.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 76; Trending score 64/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.