AutoGen — Conversational Multi-Agent Systems

import autogen

# LLM configuration
config_list = [
    {
        "model": "claude-sonnet-4-6",
        "api_key": "your-api-key",
        "base_url": "https://api.anthropic.com/v1",
        "api_type": "anthropic",
    }
]

llm_config = {
    "config_list": config_list,
    "temperature": 0,
    "timeout": 120,
}

# -------------------------------------------------------
# AGENT 1: The Assistant (LLM-powered)
# -------------------------------------------------------
assistant = autogen.AssistantAgent(
    name="DataScientist",
    system_message=(
        "You are an expert data scientist. When given a task, "
        "write clean, well-commented Python code to solve it. "
        "Always include error handling and print results. "
        "If code execution reveals an error, debug and fix it."
    ),
    llm_config=llm_config,
)

# -------------------------------------------------------
# AGENT 2: The UserProxy (executes code, optionally involves human)
# -------------------------------------------------------
user_proxy = autogen.UserProxyAgent(
    name="CodeRunner",
    human_input_mode="NEVER",  # run fully autonomously
    # human_input_mode="TERMINATE",  # ask human only before terminating
    # human_input_mode="ALWAYS",     # ask human every turn
    max_consecutive_auto_reply=10,
    is_termination_msg=lambda msg: "TASK_COMPLETE" in msg.get("content", ""),
    code_execution_config={
        "work_dir": "/tmp/autogen_workspace",
        "use_docker": False,  # set True for sandboxed execution
    },
    system_message="Execute code provided by DataScientist. Report results accurately.",
)

# -------------------------------------------------------
# Start the conversation
# -------------------------------------------------------
user_proxy.initiate_chat(
    assistant,
    message=(
        "Analyze the Iris dataset: load it from sklearn, compute summary statistics, "
        "find correlations between features, and identify which features best "
        "distinguish the three species. Print all findings."
    ),
)

import autogen

llm_config = {"config_list": config_list, "temperature": 0}

# -------------------------------------------------------
# Define specialized agents
# -------------------------------------------------------
product_manager = autogen.AssistantAgent(
    name="ProductManager",
    system_message=(
        "You are a product manager. You understand user needs, "
        "prioritize features, and define requirements. "
        "When the technical solution is ready, say 'APPROVED' to end discussion."
    ),
    llm_config=llm_config,
)

backend_dev = autogen.AssistantAgent(
    name="BackendDev",
    system_message=(
        "You are a senior backend engineer specializing in Python APIs. "
        "You propose technical solutions, write server-side code, "
        "and raise concerns about scalability and security."
    ),
    llm_config=llm_config,
)

frontend_dev = autogen.AssistantAgent(
    name="FrontendDev",
    system_message=(
        "You are a frontend engineer specializing in React. "
        "You design UI/UX, write component code, "
        "and ensure the API contract suits the frontend needs."
    ),
    llm_config=llm_config,
)

qa_engineer = autogen.AssistantAgent(
    name="QAEngineer",
    system_message=(
        "You are a QA engineer. You review all proposed code and designs, "
        "identify edge cases and potential bugs, and suggest test cases."
    ),
    llm_config=llm_config,
)

# UserProxy coordinates and can execute code
coordinator = autogen.UserProxyAgent(
    name="Coordinator",
    human_input_mode="NEVER",
    max_consecutive_auto_reply=0,
    is_termination_msg=lambda msg: "APPROVED" in msg.get("content", ""),
    code_execution_config={"work_dir": "/tmp/autogen_workspace", "use_docker": False},
)

# -------------------------------------------------------
# Set up group chat
# -------------------------------------------------------
group_chat = autogen.GroupChat(
    agents=[coordinator, product_manager, backend_dev, frontend_dev, qa_engineer],
    messages=[],
    max_round=20,
    speaker_selection_method="auto",  # let the manager decide who speaks next
)

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

# -------------------------------------------------------
# Initiate the group discussion
# -------------------------------------------------------
coordinator.initiate_chat(
    manager,
    message=(
        "Design and implement a simple REST API endpoint for user authentication "
        "(POST /auth/login) that accepts email and password, validates credentials "
        "against a mock database, and returns a JWT token. "
        "Include the frontend form component and test cases."
    ),
)

# A writer agent that uses a nested chat with a critic for self-revision
writer = autogen.AssistantAgent(
    name="Writer",
    system_message=(
        "You write technical blog posts. After writing, "
        "you will consult with a critic to improve the draft."
    ),
    llm_config=llm_config,
)

critic = autogen.AssistantAgent(
    name="Critic",
    system_message=(
        "You are a harsh but fair technical editor. "
        "Review the blog post draft and provide specific, actionable feedback "
        "on clarity, accuracy, and structure. Rate it 1-10."
    ),
    llm_config=llm_config,
)

user = autogen.UserProxyAgent(
    name="User",
    human_input_mode="NEVER",
    is_termination_msg=lambda x: x.get("content", "").find("TERMINATE") >= 0,
    code_execution_config=False,
)

# Register a nested chat: when user talks to writer,
# writer automatically consults critic before finalizing
writer.register_nested_chats(
    [{"recipient": critic, "message": "Please review this draft: {last_message}", "max_turns": 2}],
    trigger=user,
)

user.initiate_chat(
    writer,
    message="Write a 300-word intro to vector databases for a developer audience.",
)