AutoGen

Guard Microsoft AutoGen multi-agent conversations.

The AutoGen integration is in Beta. The API may change; pin your SDK version and watch the changelog.

Beta status

AutoGen support guards messages exchanged between agents so unsafe or injected content is caught mid-conversation rather than only at the end.

Install

Python
pip install veldrixai pyautogen

Usage

Register a reply hook that evaluates each agent message and blocks or annotates it according to your policy:

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

def guarded_reply(recipient, messages, sender, config):
    last = messages[-1]["content"]
    trust = veldrix.evaluate_sync(prompt="", response=last)
    if trust.verdict == "BLOCK":
        return True, "[message blocked by VeldrixAI policy]"
    return False, None

assistant.register_reply([Agent, None], guarded_reply)
Was this page helpful?