Multi-Agent Governance • Microsoft AutoGen

Security for
AutoGen
Swarms

Securing autonomous conversations from rogue code execution and unauthorized API access.

Quick Install

pip install suprawall

Implementation

# Protecting AutoGen conversations with SupraWall

import autogen
from suprawall.autogen import SupraWallMiddleware

# Initialize firewall
sw = SupraWallMiddleware(api_key="sw_live_...")

# Configure your AutoGen agents
assistant = autogen.AssistantAgent("assistant", ...)
user_proxy = autogen.UserProxyAgent("user_proxy", ...)

# Register SupraWall to intercept every interaction
sw.register(assistant)
sw.register(user_proxy)

# Conversations are now audit-trailed and policy-governed
user_proxy.initiate_chat(assistant, message="Analyze this shell script...")

Why AutoGen Needs SupraWall

AutoGen's power lies in the interaction between recursive agents. However, without a dedicated security layer, your user_proxy can easily be tricked into executing malicious code generated by an assistant model.

SupraWall intercepts the messaging layer of AutoGen. We verify the code blocks being passed between agents against your local or cloud-based policies, ensuring that even if an assistant generates a `rm -rf /` command, it is blocked before the user_proxy executes it.

Code Sandboxing

Real-time inspection of Python/Bash code blocks.

Recursive Depth Guard

Prevent infinite message loops from draining model budget.

Identity Governance

Ensure agents correctly represent their assigned roles.

EU AI Act Compliance

Automatic logging and oversight for Article 12 accountability.

Secure Your
Conversations.

Start Building for Free