Library / SDK
Guardrails & Structure
Rebuff
1Detect 2and 3prevent 4LLM 5prompt 6injection
the six Ws · specification
W1
Who
Maintained by Protect AI (the Rebuff project).
W2
What
A self-hardening prompt-injection detector combining heuristics, an LLM check, a vector DB of attacks, and canary tokens.
W3
Where
A Python/JS SDK plus a self-hostable server; also a hosted API.
W4
When
Reach for it when you need a defense layer against prompt injection.
W5
Why
Layers multiple detection strategies and learns from attacks to cut risk.
W6
With
Uses an LLM for detection and a vector store for attack signatures.
W7
Watch
Maintained by Protect AI; open-source (Apache-2.0) and self-hostable. Self-hosted detection can still call external LLM and vector-store APIs.
alternatives