Group Chat with Tools

This user guide explains how to set up a group chat where each agent has unique capabilities and is equipped with specialized tools to perform specific tasks.

Installation

pip install -U ag2[openai]

Note: If you have been using autogen or pyautogen, all you need to do is upgrade it using:

pip install -U autogen[openai]

or

pip install -U pyautogen[openai]

as pyautogen, autogen, and ag2 are aliases for the same PyPI package.

Imports

import os
import random

from autogen import (
    ConversableAgent,
    GroupChat,
    GroupChatManager,
    UserProxyAgent,
    register_function,
    LLMConfig,
)

Agent Configuration

The GroupChat will contain three agents:

• sales_agent - Responsible for selling tickets.
• cancellation_agent - Handles ticket cancellations.
• user_proxy - Acts as an intermediary between the user and other agents.

llm_config = LLMConfig(
    api_type="openai",
    model="gpt-4o-mini",
    api_key=os.environ["OPENAI_API_KEY"],
)

with llm_config:
    sales_agent = ConversableAgent(name="SalesAgent")

    cancellation_agent = ConversableAgent(name="CanelationAgent")

user_proxy = UserProxyAgent(
    name="user_proxy",
    human_input_mode="ALWAYS",
    code_execution_config={
        "use_docker": False,
    },  # Please set use_docker=True if Docker is available to run the generated code. Using Docker is safer and recommended over running the generated code directly on the host machine.
)

Tools Registration

In AG2, tool usage follows two steps:

• An agent suggests a tool to use (via its LLM).
• Another agent executes the tool.

We will define two tools:

• buy_airplane_ticket: Suggested by sales_agent and executed by user_proxy after user verification.
• cancel_airplane_ticket: Suggested by cancellation_agent and executed by user_proxy after user verification.

def buy_airplane_ticket(from_location: str, to_location: str, date: str) -> str:
    ticket_number = random.randint(1000, 9999)
    return f"""Your ticket from {from_location} to {to_location} on {date} has been booked.
Your ticket number is {ticket_number}.
Please keep this number for future reference.
"""

register_function(
    buy_airplane_ticket,
    caller=sales_agent,
    executor=user_proxy,
    description="Buy an airplane ticket",
)

def cancel_airplane_ticket(ticket_number: str) -> str:
    return f"Your ticket with ticket number {ticket_number} has been canceled"

register_function(
    cancel_airplane_ticket,
    caller=cancellation_agent,
    executor=user_proxy,
    description="Cancel an airplane ticket",
)

Creating and Initiating the Group Chat

Now, let’s create and start the GroupChat with the three agents.

groupchat = GroupChat(
    agents=[user_proxy, cancellation_agent, sales_agent],
    speaker_selection_method="auto",
    messages=[],
)

manager = GroupChatManager(
    name="group_manager",
    groupchat=groupchat,
    llm_config=llm_config,
)

user_proxy.initiate_chat(
    recipient=manager,
    message="I need to buy a plane ticket from New York to Los Angeles on 12th of April 2025",
)

Example Console Output

user_proxy (to group_manager):

I need to buy a plane ticket from New York to Los Angeles on 12th of April 2025

--------------------------------------------------------------------------------

Next speaker: SalesAgent

>>>>>>>> USING AUTO REPLY...
SalesAgent (to group_manager):

***** Suggested tool call (call_H7AqjaWAohk3zuezgPmvoxjB): buy_airplane_ticket *****
Arguments:
{"from_location":"New York","to_location":"Los Angeles","date":"2025-04-12"}
************************************************************************************

--------------------------------------------------------------------------------

Next speaker: user_proxy

>>>>>>>> NO HUMAN INPUT RECEIVED.

>>>>>>>> USING AUTO REPLY...

>>>>>>>> EXECUTING FUNCTION buy_airplane_ticket...
Call ID: call_H7AqjaWAohk3zuezgPmvoxjB
Input arguments: {'from_location': 'New York', 'to_location': 'Los Angeles', 'date': '2025-04-12'}
user_proxy (to group_manager):

***** Response from calling tool (call_H7AqjaWAohk3zuezgPmvoxjB) *****
Your ticket from New York to Los Angeles on 2025-04-12 has been booked.
Your ticket number is 6841.
Please keep this number for future reference.

**********************************************************************

--------------------------------------------------------------------------------

Next speaker: CanelationAgent

>>>>>>>> USING AUTO REPLY...
CanelationAgent (to group_manager):

Your ticket from New York to Los Angeles on April 12, 2025, has been successfully booked. Your ticket number is **6841**. Please keep this number for future reference.

--------------------------------------------------------------------------------

Next speaker: CanelationAgent

>>>>>>>> USING AUTO REPLY...
CanelationAgent (to group_manager):

Your ticket from New York to Los Angeles on April 12, 2025, has been successfully booked. Your ticket number is **6841**. Please keep this number for future reference.

--------------------------------------------------------------------------------

Next speaker: user_proxy

user_proxy (to group_manager):

Cancel it

--------------------------------------------------------------------------------

Next speaker: CanelationAgent

>>>>>>>> USING AUTO REPLY...
CanelationAgent (to group_manager):

***** Suggested tool call (call_pSxE8W5g0HaAZ9UFOrbgD9sx): cancel_flight_ticket *****
Arguments:
{"ticket_number":"6841"}
*************************************************************************************

--------------------------------------------------------------------------------

Next speaker: user_proxy

>>>>>>>> NO HUMAN INPUT RECEIVED.

>>>>>>>> USING AUTO REPLY...

>>>>>>>> EXECUTING FUNCTION cancel_flight_ticket...
Call ID: call_pSxE8W5g0HaAZ9UFOrbgD9sx
Input arguments: {'ticket_number': '6841'}
user_proxy (to group_manager):

***** Response from calling tool (call_pSxE8W5g0HaAZ9UFOrbgD9sx) *****
Your ticket with ticket number 6841 has been canceled
**********************************************************************

--------------------------------------------------------------------------------

Next speaker: user_proxy

user_proxy (to group_manager):

thanks

--------------------------------------------------------------------------------

Edit this page