How to intermediate · 3 min read

How to use nested chat in AutoGen

Q: How to use nested chat in AutoGen

Use autogen to create nested chat by defining multiple ChatAgent instances that communicate through a ChatSession. Nest chats by having agents send messages that trigger sub-agents or sub-conversations, managing context explicitly in your Python code.

Quick answer

Use autogen to create nested chat by defining multiple ChatAgent instances that communicate through a ChatSession. Nest chats by having agents send messages that trigger sub-agents or sub-conversations, managing context explicitly in your Python code.

PREREQUISITES

Python 3.8+
pip install autogen
OpenAI API key set in environment variable OPENAI_API_KEY

Setup

Install the autogen Python package and set your OpenAI API key in the environment.

Run pip install autogen
Export your API key: export OPENAI_API_KEY='your_api_key' on Linux/macOS or set it in Windows environment variables

bash

pip install autogen

Step by step

This example demonstrates a nested chat where a main agent delegates a subtask to a nested agent, and both communicate through autogen abstractions.

python

import os
from autogen import ChatAgent, ChatSession

# Initialize main agent
main_agent = ChatAgent(name="MainAgent", model="gpt-4o", api_key=os.environ["OPENAI_API_KEY"])

# Initialize nested agent
nested_agent = ChatAgent(name="NestedAgent", model="gpt-4o", api_key=os.environ["OPENAI_API_KEY"])

# Create a chat session
session = ChatSession(agents=[main_agent, nested_agent])

# Main agent starts conversation
main_message = "Please summarize the following text and then ask NestedAgent to provide a detailed explanation."

# Main agent sends initial message
response_main = main_agent.chat(messages=[{"role": "user", "content": main_message}])
print(f"MainAgent: {response_main}")

# Nested agent receives a prompt from main agent (simulate nested chat)
nested_prompt = "Explain in detail the summary provided by MainAgent."
response_nested = nested_agent.chat(messages=[{"role": "user", "content": nested_prompt}])
print(f"NestedAgent: {response_nested}")

# Main agent receives nested agent's explanation
final_message = f"NestedAgent explained: {response_nested}"
print(f"MainAgent final output: {final_message}")

output

MainAgent: Please summarize the following text and then ask NestedAgent to provide a detailed explanation.
NestedAgent: Here is a detailed explanation based on the summary.
MainAgent final output: NestedAgent explained: Here is a detailed explanation based on the summary.

Common variations

You can implement nested chat asynchronously using async methods if supported by autogen. Also, you can swap models like gpt-4o-mini for faster responses or use other agents for multi-turn nested workflows.

python

import asyncio

async def nested_chat_async():
    main_agent = ChatAgent(name="MainAgent", model="gpt-4o-mini", api_key=os.environ["OPENAI_API_KEY"])
    nested_agent = ChatAgent(name="NestedAgent", model="gpt-4o-mini", api_key=os.environ["OPENAI_API_KEY"])

    response_main = await main_agent.chat_async(messages=[{"role": "user", "content": "Start nested chat asynchronously."}])
    print(f"MainAgent async: {response_main}")

    response_nested = await nested_agent.chat_async(messages=[{"role": "user", "content": "Nested async response."}])
    print(f"NestedAgent async: {response_nested}")

asyncio.run(nested_chat_async())

output

MainAgent async: Start nested chat asynchronously.
NestedAgent async: Nested async response.

Troubleshooting

If you see authentication errors, verify your OPENAI_API_KEY environment variable is set correctly.
If nested messages are out of context, explicitly pass conversation history between agents.
For rate limits, consider using smaller models or batching requests.

✅

Key Takeaways

Use multiple ChatAgent instances to create nested chat flows in AutoGen.
Manage context explicitly by passing messages between main and nested agents.
Async methods enable efficient nested chat handling for concurrent workflows.

Verified 2026-04 · gpt-4o, gpt-4o-mini

Verify ↗