Library / SDK
Guardrails & Structure
LLM Guard
1Security 2toolkit 3sanitizing 4LLM 5prompt 6inputs
the six Ws · specification
W1
Who
Maintained by Protect AI.
W2
What
A security toolkit with input/output scanners for injection, toxicity, PII, secrets, and bias.
W3
Where
A Python library, self-hosted, optionally as an API service.
W4
When
Reach for it when sanitizing prompts and completions around LLM calls.
W5
Why
Bundles many ready-made scanners so you harden LLM I/O without building each.
W6
With
Uses Hugging Face transformer models and Presidio for detection.
W7
Watch
Maintained by Protect AI; open-source (MIT) and fully self-hostable. Runs locally with local models by default, so prompts need not leave your infrastructure.
alternatives