Skip to main content

​
AutoGen Integration

Use KoreShield as a proxy for AutoGen agent requests to enforce sanitization, detection, and policy controls before traffic reaches your provider.

​
Use Cases

  • Multi-agent workflows with strict safety policies
  • Centralized auditing for agent traffic
  • Shared rate limiting across agent fleets

​
Prerequisites

1

Running KoreShield Instance

Ensure you have a running KoreShield instance accessible from your application.
2

Provider API Key

Configure your provider API key on the KoreShield server.
3

Install AutoGen

pip install pyautogen

​
Environment Variables

KORESHIELD_BASE_URL=http://localhost:8000
KORESHIELD_API_KEY=your-koreshield-api-key

​
Configuration

​
Basic Setup

import autogen

llm_config = {
    "config_list": [
        {
            "model": "gpt-5-mini",
            "api_key": "unused",
            "base_url": "http://localhost:8000",
            "default_headers": {
                "Authorization": "Bearer your-koreshield-api-key"
            }
        }
    ],
    "temperature": 0.2
}

assistant = autogen.AssistantAgent(
    name="assistant",
    llm_config=llm_config
)

user_proxy = autogen.UserProxyAgent(
    name="user",
    human_input_mode="NEVER",
    max_consecutive_auto_reply=3
)

user_proxy.initiate_chat(
    assistant,
    message="Summarize the incident report and flag risky items."
)

​
Multi-Agent Workflows

​
Secure Agent Fleet

import autogen
import os

# KoreShield configuration
koreshield_config = {
    "base_url": os.getenv("KORESHIELD_BASE_URL", "http://localhost:8000"),
    "api_key": os.getenv("KORESHIELD_API_KEY")
}

llm_config = {
    "config_list": [
        {
            "model": "gpt-4",
            "api_key": "unused",  # Actual key stored on KoreShield server
            "base_url": koreshield_config["base_url"],
            "default_headers": {
                "Authorization": f"Bearer {koreshield_config['api_key']}"
            }
        }
    ],
    "temperature": 0.7
}

# Create multiple agents
researcher = autogen.AssistantAgent(
    name="researcher",
    system_message="You are a research assistant.",
    llm_config=llm_config
)

writer = autogen.AssistantAgent(
    name="writer",
    system_message="You are a technical writer.",
    llm_config=llm_config
)

critic = autogen.AssistantAgent(
    name="critic",
    system_message="You are a critical reviewer.",
    llm_config=llm_config
)

user_proxy = autogen.UserProxyAgent(
    name="user",
    human_input_mode="TERMINATE",
    max_consecutive_auto_reply=10
)

# All agent interactions are protected by KoreShield
groupchat = autogen.GroupChat(
    agents=[user_proxy, researcher, writer, critic],
    messages=[],
    max_round=12
)

manager = autogen.GroupChatManager(
    groupchat=groupchat,
    llm_config=llm_config
)

user_proxy.initiate_chat(
    manager,
    message="Research and write a report on LLM security best practices."
)

​
Custom Security Policies

​
Per-Agent Configuration

# High security for customer-facing agent
customer_agent_config = {
    "config_list": [{
        "model": "gpt-4",
        "base_url": "http://localhost:8000",
        "default_headers": {
            "Authorization": f"Bearer {os.getenv('KORESHIELD_API_KEY')}",
            "X-Security-Level": "high"
        }
    }],
    "temperature": 0.3
}

# Lower security for internal analysis agent
internal_agent_config = {
    "config_list": [{
        "model": "gpt-4",
        "base_url": "http://localhost:8000",
        "default_headers": {
            "Authorization": f"Bearer {os.getenv('KORESHIELD_API_KEY')}",
            "X-Security-Level": "medium"
        }
    }],
    "temperature": 0.7
}

customer_agent = autogen.AssistantAgent(
    name="customer_support",
    llm_config=customer_agent_config
)

analyst_agent = autogen.AssistantAgent(
    name="data_analyst",
    llm_config=internal_agent_config
)

​
Error Handling

import autogen
from typing import Dict, Any

def safe_chat(agent: autogen.AssistantAgent, message: str) -> Dict[str, Any]:
    """Execute chat with error handling for security blocks"""
    try:
        user_proxy = autogen.UserProxyAgent(
            name="user",
            human_input_mode="NEVER",
            max_consecutive_auto_reply=1
        )
        
        user_proxy.initiate_chat(agent, message=message)
        
        return {
            "success": True,
            "response": user_proxy.last_message()["content"]
        }
        
    except Exception as e:
        if "403" in str(e) or "Blocked" in str(e):
            return {
                "success": False,
                "error": "security_violation",
                "message": "Request blocked by security policy"
            }
        else:
            return {
                "success": False,
                "error": "unknown",
                "message": str(e)
            }

# Usage
result = safe_chat(assistant, "Process this user input")
if not result["success"]:
    print(f"Error: {result['message']}")

​
Security Notes

Important Security Practices
  • Keep provider API keys on the KoreShield server only
  • Use KoreShield API key for client access
  • Configure policies in your KoreShield dashboard
  • Enable audit logging for compliance
Store provider API keys (OpenAI, Anthropic, etc.) on the KoreShield server, not in your application code. The application only needs the KoreShield API key.
Ensure KoreShield is accessible from your AutoGen application but not exposed to the public internet unless properly secured.
KoreShield provides centralized rate limiting across all agents to prevent abuse and manage costs.

​
Troubleshooting

Confirm the Authorization: Bearer <KORESHIELD_API_KEY> header is correctly set in your llm_config.
Verify that AutoGen is using the proxy base_url instead of the default OpenAI endpoint.
Check that provider API keys are correctly configured on the KoreShield server.

​
Next Steps

Python SDK

Review SDK usage and examples

Configuration

Configure providers and security policies

AutoGen Docs

Official AutoGen documentation

Build docs developers (and LLMs) love

Get started for freeTalk to us