autogen-omega 0.3.0

OmegaEngine governance integration for autogen-omega

autogen-omega

Microsoft AutoGen Integration for OmegaEngine

Documentation • Playground • Issues

---

✨ Features

• 🤖 Agent Wrapper — Add governance to any AutoGen agent
• 🔄 Conversation Guard — Monitor multi-agent dialogues
• 🛡️ Tool Execution Control — Validate tool calls before execution
• 📊 Group Chat Monitoring — Oversee entire agent teams
• ⚡ Human-in-the-Loop — Automatic escalation on policy triggers

---

📦 Installation

pip install autogen-omega

---

🚀 Quick Start

Governed Assistant Agent

from autogen import AssistantAgent, UserProxyAgent
from autogen_omega import OmegaAssistantAgent

# Create a governed agent
assistant = OmegaAssistantAgent(
    name="analyst",
    system_message="You are a financial analyst.",
    omega_api_key="your-omega-api-key",
    policy_id="finance_policy",
    llm_config={"model": "gpt-4"}
)

user_proxy = UserProxyAgent(
    name="user",
    human_input_mode="NEVER",
)

# Start governed conversation
user_proxy.initiate_chat(
    assistant,
    message="Analyze this investment opportunity..."
)

Group Chat with Governance

from autogen import GroupChat, GroupChatManager
from autogen_omega import OmegaGroupChatManager

# Create your agents
researcher = OmegaAssistantAgent(name="researcher", ...)
analyst = OmegaAssistantAgent(name="analyst", ...)
writer = OmegaAssistantAgent(name="writer", ...)

# Governed group chat
group_chat = GroupChat(
    agents=[researcher, analyst, writer],
    messages=[],
)

manager = OmegaGroupChatManager(
    groupchat=group_chat,
    omega_api_key="your-omega-api-key",
    policy_id="team_policy",
)

# All agent interactions are monitored
user_proxy.initiate_chat(manager, message="Research and write a report...")

---

🛡️ Tool Execution Control

Validate tool/function calls before execution:

from autogen_omega import OmegaToolValidator

validator = OmegaToolValidator(
    omega_api_key="your-omega-api-key",
    policy_id="tool_safety",
)

@validator.guard
def execute_sql(query: str) -> str:
    """Execute SQL query - governed by OmegaEngine"""
    return db.execute(query)

# SQL queries are validated before execution
# Dangerous queries like DROP TABLE are blocked

---

📊 Conversation Auditing

Every agent message is logged:

from autogen_omega import OmegaConversationLogger

logger = OmegaConversationLogger(
    omega_api_key="your-omega-api-key",
    session_id="session_123",
)

# Attach to agents
assistant.register_hook("on_message", logger.log)

# Access full conversation audit
audit = logger.get_audit()
print(f"Messages: {audit.message_count}")
print(f"Policy violations: {audit.violations}")

---

🎯 Use Cases

Use Case	How It Helps
Research Teams	Ensure agents stay on topic
Code Generation	Prevent malicious code execution
Document Processing	Protect sensitive data
Customer Service	Enforce response policies

---

📄 License

Licensed under the Apache License 2.0.

---

Built with ❤️ by the OmegaEngine team