CrewAI

Guard CrewAI agents and tasks so every step in a crew is evaluated.

Install

Python
pip install veldrixai crewai

Guarding tasks

Wrap the callable a CrewAI agent uses to produce output with the guard decorator, or evaluate task results explicitly before passing them downstream:

Python
from veldrixai import Veldrix
veldrix = Veldrix(api_key="vx-live-...")

result = task.execute()
trust = veldrix.evaluate_sync(prompt=task.description, response=str(result))
if trust.verdict == "BLOCK":
    result = "[blocked by policy]"

Notes

Evaluate the output of each agent in a multi-agent crew, especially any agent that consumes external tool or web data — that is where indirect injection enters.

Was this page helpful?