Welcome to the Autogen-MagenticOne (AGMO) repository! 🚀
This project showcases MagenticOne's multiagent features to solve complex problems across a variety of domains.
- 🌟 Project Overview
- 🛠️ Pre-requisites
- 📖 Introduction to AGMO
- 🧩 Components of MagenticOne
- 💻 Example Code
⚠️ Important Notes- 📜 License
Autogen-MagenticOne (AGMO) is a multiagent system designed to resolve open-ended tasks. It leverages cutting-edge tools such as SemanticKernel, Gradio, and Azure OpenAI to create an advanced architecture for agent-based problem-solving.
To get started, you'll need the following:
-
SemanticKernel
Refer to my repository on Agent Orchestration for more details:
🔗 Microsoft-Semantic-Kernel-for-Agentic-AI -
Gradio
Gradio helps in building a stunning and interactive chatbot interface. 🤖 -
Azure OpenAI Compatibility
This project is crafted exclusively for Azure OpenAI subscribers. Features showcased here rely heavily on Azure OpenAI components.
With the rapid evolution of AI frameworks, we now have Autogen from Microsoft, alongside popular tools like LangChain, LLamaIndex, and Semantic Kernel.
Autogen introduces Agent Chat, and one of its advanced concepts is MagenticOne, a multiagent system designed for solving open-ended tasks.
MagenticOne is orchestrated by the GroupChat Orchestrator and comprises the following agents:
- Functions as a web surfer capable of searching the internet and navigating web pages.
- Uses Chromium browser via Playwright for interaction.
- Requires a multimodal model client (e.g., GPT-4o) for optimal functionality.
Using MultimodalWebSurfer carries inherent risks, such as:
- Interacting with prompt injection attacks from webpages.
- Recruiting humans or accepting cookie agreements without supervision.
Always monitor and operate in a controlled environment.
- Acts as a local file navigator.
- Can traverse the file system, open, and read common file types and hierarchies.
- Provides coding assistance using an LLM model client.
- Generates code based on user instructions provided by the orchestrator.
- Orchestrates the agents to create plans and action items.
- Monitors progress towards goals and dynamically adjusts plans if results deviate or fail.
- Generates alternative solutions until optimized results are achieved.
The above agents are available as individual task agents as well in the Autogen Library which can be added as a customized group of agents in the MagenticOneGroupChat orchestrator.
Here’s how to use MagenticOneGroupChat with minimal configuration:
from autogen import OpenAIChatCompletionClient, MultimodalWebSurfer, MagenticOneGroupChat
# Initialize the model client
model_client = OpenAIChatCompletionClient(model="gpt-4o")
# Create a MultimodalWebSurfer agent
surfer = MultimodalWebSurfer(
"WebSurfer",
model_client=model_client,
)
# GroupChat orchestrator with the surfer agent
team = MagenticOneGroupChat([surfer], model_client=model_client) However, the MagenticOne helper class bundles all agents together, allowing you to focus on tasks without worrying about individual configurations which is explained below
Using MagenticOne involves interacting with a digital world designed for humans, which carries inherent risks. To minimize these risks, follow these precautions:
- Containers: Run all tasks in Docker containers to isolate agents and prevent direct system attacks. 🐳
- Virtual Environment: Use a virtual environment to prevent agents from accessing sensitive data.
- Monitor Logs: Closely monitor logs during and after execution to detect risky behavior. 📊
- Human Oversight: Operate with a human in the loop to supervise agents and prevent unintended consequences. 👀
- Limit Access: Restrict agents’ access to the internet and sensitive resources.
- Safeguard Data: Ensure agents do not have access to sensitive information.
Agents may attempt risky actions like recruiting humans for help or accepting cookie agreements without human involvement. Always ensure agents operate in a controlled environment.
Follow the instructions in the youtube link on the setup and execution.
This project is licensed under the MIT License. 📄
Special thanks to Microsoft for creating the Autogen framework and advancing agent-based architectures.