Stage 10 — Microsoft AutoGen

Microsoft AutoGen  ·  Comprehensive Technical Training  ·  ⏱ 8–10 hours

Learning Objectives

By the end of this stage you will be able to:

• Build agents with Microsoft AutoGen framework
• Implement agent conversations and message passing
• Use AutoGen Core for distributed agent systems
• Handle gRPC-based agent communication
• Design scalable multi-agent architectures

Section 1: AutoGen Basics

AutoGen enables automatic message-based conversations between agents. Agents pass messages back and forth until reaching an end state.

Simple Two-Agent Chat

from autogen import ConversableAgent

# Define agents
user_proxy = ConversableAgent(
    name="user_proxy",
    system_message="You are a helpful user...",
    llm_config={"model": "gpt-4o"},
)

assistant = ConversableAgent(
    name="assistant",
    system_message="You are a helpful assistant...",
    llm_config={"model": "gpt-4o"},
)

# Start conversation
user_proxy.initiate_chat(
    assistant,
    message="Help me debug this code",
    max_consecutive_auto_reply=5,
)
# Agents exchange messages automatically until stopping condition

Section 2: Tool Definition in AutoGen

tools = [
    {
        "type": "function",
        "function": {
            "name": "execute_code",
            "description": "Execute Python code",
            "parameters": {
                "type": "object",
                "properties": {"code": {"type": "string"}},
            },
        },
    }
]

# Agent with tools
coder = ConversableAgent(
    name="coder",
    llm_config={"model": "gpt-4o", "tools": tools},
)

critic = ConversableAgent(
    name="critic",
    llm_config={"model": "gpt-4o"},
)

# Coder solves problem, critic reviews
coder.initiate_chat(critic, message="Write a function to...")

Section 3: AutoGen Core and Distributed Agents

AutoGen Core supports distributed agent systems using gRPC for communication.

from autogen_core import DefaultMessageHandler, DefaultRuntime
from autogen_agentchat.agents import CodeExecutorAgent, AssistantAgent

# Define runtime with gRPC
runtime = DefaultRuntime(
    grpc_server_host="127.0.0.1",
    grpc_server_port=50051,
)

# Create agents registered with runtime
executor = CodeExecutorAgent(
    runtime=runtime,
    name="executor",
)

assistant = AssistantAgent(
    runtime=runtime,
    name="assistant",
    llm_config={"model": "gpt-4o"},
)

# Agents communicate via gRPC
async def main():
    # Send message through runtime
    await runtime.send_message("executor", "Execute this code...")

asyncio.run(main())

What's Next

Stage 11 introduces the Model Context Protocol (MCP), a standard for agents to discover and use external tools and services seamlessly.