4 Steps to Construct Multi-Agent Nested Chats with AutoGen

Chatbots have advanced exponentially with developments in synthetic intelligence (AI). Now, with the onset of AI brokers, they’ve grow to be able to dealing with extra complicated and layered interactions, far past conventional conversational limits. In our earlier article on Constructing Multi-Agent Chatbots with AutoGen, we explored the idea of sequential chat utilizing AutoGen, which permits structured, turn-based communication amongst a number of brokers. Now, constructing upon that basis, we’ll flip to a extra complicated function: nested chat. With AutoGen’s strong framework, nested conversations allow bots to keep up fluid exchanges as a substitute of following a set sequence. They’ll delve into different instruments, deal with interruptions, and resume easily, all inside a single dialog circulation. This text will information you thru implementing nested chat in AutoGen and spotlight its relevance in creating responsive, dynamic agentic interactions.

What’s Nested Chat?

Let’s start by understanding what a nested chat is.

Think about a three-agent chat the place it’s required for 2 brokers to repeatedly speak to one another in a loop. This chat between two brokers will be added in a nested chat. As soon as this separate dialog is completed, the agent can convey again the context to the principle dialog.

The under determine exhibits the conversion circulation of a nested chat.

When the incoming message triggers a situation, that message goes to the nested chat. That nested is usually a two-agent chat, a sequential chat, or some other. Then the chat outcomes of the nested chat are despatched again to the principle dialog.

Implementing Nested Chat in AutoGen

On this article, we’ll construct an article-writing system utilizing a nested chat. For this, we’ll create three brokers – one for writing an overview of the article, one other for writing the article based mostly on this define, and one other for reviewing the article. We’ll need the author and reviewer to speak to one another a number of instances so we’ll these two in a nested chat.

Moreover, we may also present the define agent entry to a software to question the online.

Now, let’s implement this with code.

Pre-requisites

Earlier than constructing AutoGen brokers, guarantee you may have the required API keys for the required LLMs. For this train, we may also be utilizing Tavily to look the online.

Load the .env file with the API keys wanted. Right here we’ll use OpenAI and Tavily API keys ().

from dotenv import load_dotenv

load_dotenv('/house/santhosh/Initiatives/programs/Pinnacle/.env')

Outline the LLM for use as a config_list

config_list = {

	"config_list": [{"model": "gpt-4o-mini", "temperature": 0.2}]

}

Key Libraries Required

autogen-agentchat – 0.2.37

Tavily-python – 0.5.0

Now, let’s get to the implementation.

Step 1: Outline the Define Agent with the Software Use

Outline the user_proxy agent which may also execute the software. Then outline the Define Agent utilizing the LLM to generate the article define.

from autogen import ConversableAgent
user_proxy = ConversableAgent(
	title="Person",
	llm_config=False,
	is_termination_msg=lambda msg: msg.get("content material") will not be None and "TERMINATE" in msg["content"],
	human_input_mode="TERMINATE")

define = ConversableAgent(
	title="Article_outline",
	system_message="""You're an knowledgeable content material strategist tasked with creating an in depth define
                	for an article on a specified matter. Your aim is to arrange the article into
                	logical sections that assist convey the principle concepts clearly and successfully.
                	Use the web_search software if wanted.
                	Return 'TERMINATE' when the duty is completed.""",
	llm_config=config_list,
	silent=False,
)

Outline the web_search operate to question the online.

def web_search(question: str) -> str:
	tavily_client = TavilyClient()
	response = tavily_client.search(question, max_results=3, days=10, include_raw_content=True)
	return response['results']

Register the web_search operate to the define agent with the executor as user_proxy.

We’re making the executor as user_proxy in order that we will assessment the define that goes to the author agent.

register_function(
	web_search,
	caller=define,  # The assistant agent can counsel calls.
	executor=user_proxy,  # The person proxy agent can execute the calls.
	title="web_search",  # By default, the operate title is used because the software title.
	description="Searches web to get the outcomes a for given question",  # An outline of the software.
)

Step 2: Outline the Author and Reviewer Brokers

Outline one agent to generate the article content material, and one other to assessment the article and supply ideas to enhance.

author = ConversableAgent(
	title="Article_Writer",
	system_message="""You're a expert author assigned to create a complete, participating article
                	based mostly on a given define. Your aim is to comply with the construction supplied within the define,
                	increasing on every part with well-researched, clear, and informative content material.
                	Hold the article size to round 500 phrases.
                	Use the web_search software if wanted.
                	Return 'TERMINATE' when the duty is completed.""",
	llm_config=config_list,
	silent=False,
)
reviewer = ConversableAgent(
	title="Article_Reviewer",
	system_message="""You're a expert article reviewer who can assessment technical articles.
                	Assessment the given article and supply ideas to make the article extra participating and fascinating.
                	""",
	llm_config=config_list,
	silent=False,
)

Step 3: Register the Nested Chat

We will now register the nested chats for each brokers.

author.register_nested_chats(
	set off=user_proxy,
	chat_queue=[
    	{
        	"sender": reviewer,
        	"recipient": writer,
        	"summary_method": "last_msg",
        	"max_turns": 2,
    	}
	],
)

Within the above code, when user_proxy sends any message to the author agent, this can set off the nested chat. Then the author agent will write the article and the reviewer agent will assessment the article as many instances as max_turns, two on this case. Lastly, the results of the nested chat is distributed again to the person agent.

Step 4: Provoke the Nested Chat

Now that all the pieces is ready, let’s provoke the chat

chat_results = user_proxy.initiate_chats(

              	[{"recipient": outline,

                	"message": "Write an article on Magentic-One agentic system released by Microsoft.",

                	"summary_method": "last_msg",

                	},

               	{"recipient": writer,

                	"message": "This is the article outline",

                	"summary_method": "last_msg",

                	}])

Right here, we’re going to write an article in regards to the Magentic-One agentic system. First, the user_proxy agent will provoke a chat with the Define Agent, after which with the Author Agent.

Now, the output of the above code shall be like this:

As we will see, the user_proxy first sends a message stating the subject of the article to the Define Agent. This triggers the software name and user_proxy executes the software. Based mostly on these outcomes, the Define Agent will generate the define and ship it to the author agent. After that, the nested chat between the author agent and reviewer agent continues as mentioned above.

Now, let’s print the ultimate consequence, which is the article about magentic-one.

print(chat_results[1].chat_history[-2]['content'])

Conclusion

Nested chat in AutoGen enhances chatbot capabilities by enabling complicated, multitasking interactions inside a single dialog circulation. Nested chat permits bots to provoke separate, specialised chats and combine their outputs seamlessly. This function helps dynamic, focused responses throughout numerous functions, from e-commerce to healthcare. With nested chat, AutoGen paves the best way for extra responsive, context-aware AI methods. This allows builders to construct refined chatbots that meet numerous person wants effectively.

If you wish to study extra about AI Brokers, checkout our unique Agentic AI Pioneer Program!

I’m working as an Affiliate Information Scientist at Analytics Vidhya, a platform devoted to constructing the Information Science ecosystem. My pursuits lie within the fields of Pure Language Processing (NLP), Deep Studying, and AI Brokers.