Check a prompt before you send it.

Install the ferroguard package. Import llm_guardrails. The library redacts some PII, scores a few injection patterns, and runs a small set of output checks. It does not call a model.

pip install ferroguard

The import name is llm_guardrails, not ferroguard.

quickstart.py

from llm_guardrails import GuardrailsPipeline

pipeline = GuardrailsPipeline()
pre = pipeline.pre_process("Mail a@b.co about the notes.")
print(pre.sanitised_text)
print("blocked", pre.blocked)

Mail <<EMAIL_1>> about the notes.
blocked False

Guards · What the checks leave out