Introduction to AutoGen

Welcome! AutoGen is an open-source framework that leverages multipleagents to enable complex workflows. This tutorial introduces basicconcepts and building blocks of AutoGen.

Why AutoGen?​

The whole is greater than the sum of its parts.
-Aristotle

While there are many definitions of agents, in AutoGen, an agent is anentity that can send messages, receive messages and generate a replyusing models, tools, human inputs or a mixture of them. This abstractionnot only allows agents to model real-world and abstract entities, suchas people and algorithms, but it also simplifies implementation ofcomplex workflows as collaboration among agents.

Further, AutoGen is extensible and composable: you can extend a simpleagent with customizable components and create workflows that can combinethese agents and power a more sophisticated agent, resulting inimplementations that are modular and easy to maintain.

Most importantly, AutoGen is developed by a vibrant community ofresearchers and engineers. It incorporates the latest research inmulti-agent systems and has been used in many real-world applications,including agent platform, advertising, AI employees, blog/articlewriting, blockchain, calculate burned areas by wildfires, customersupport, cybersecurity, data analytics, debate, education, finance,gaming, legal consultation, research, robotics, sales/marketing, socialsimulation, software engineering, software security, supply chain,t-shirt design, training data generation, Youtube service…

Installation​

The simplest way to install AutoGen is from pip:pip install autogen.Find more options inInstallation.

Agents​

In AutoGen, an agent is an entity that can send and receive messages toand from other agents in its environment. An agent can be powered bymodels (such as a large language model like GPT-4), code executors (suchas an IPython kernel), human, or a combination of these and otherpluggable and customizable components.

An example of such agents is the built-inConversableAgent whichsupports the following components:

1. A list of LLMs
2. A code executor
3. A function and tool executor
4. A component for keeping human-in-the-loop

You can switch each component on or off and customize it to suit theneed of your application. For advanced users, you can add additionalcomponents to the agent by usingregistered_reply.

LLMs, for example, enable agents to converse in natural languages andtransform between structured and unstructured text. The followingexample shows aConversableAgent with a GPT-4 LLM switched on andother components switched off:

import os

from autogenimport ConversableAgent

agent= ConversableAgent(
"chatbot",
    llm_config={"config_list":[{"model":"gpt-4","api_key": os.environ.get("OPENAI_API_KEY")}]},
    code_execution_config=False,# Turn off code execution, by default it is off.
    function_map=None,# No registered functions, by default it is None.
    human_input_mode="NEVER",# Never ask for human input.
)

Thellm_config argument contains a list of configurations for theLLMs. SeeLLM Configuration formore details.

You can ask this agent to generate a response to a question using thegenerate_reply method:

reply= agent.generate_reply(messages=[{"content":"Tell me a joke.","role":"user"}])
print(reply)

Sure, here's a light-hearted joke for you:

Why don't scientists trust atoms?

Because they make up everything!

Roles and Conversations​

In AutoGen, you can assign roles to agents and have them participate inconversations or chat with each other. A conversation is a sequence ofmessages exchanged between agents. You can then use these conversationsto make progress on a task. For example, in the example below, we assigndifferent roles to two agents by setting theirsystem_message.

cathy= ConversableAgent(
"cathy",
    system_message="Your name is Cathy and you are a part of a duo of comedians.",
    llm_config={"config_list":[{"model":"gpt-4","temperature":0.9,"api_key": os.environ.get("OPENAI_API_KEY")}]},
    human_input_mode="NEVER",# Never ask for human input.
)

joe= ConversableAgent(
"joe",
    system_message="Your name is Joe and you are a part of a duo of comedians.",
    llm_config={"config_list":[{"model":"gpt-4","temperature":0.7,"api_key": os.environ.get("OPENAI_API_KEY")}]},
    human_input_mode="NEVER",# Never ask for human input.
)

Now that we have two comedian agents, we can ask them to start a comedyshow. This can be done using theinitiate_chat method. We set themax_turns to 2 to keep the conversation short.

result= joe.initiate_chat(cathy, message="Cathy, tell me a joke.", max_turns=2)

joe(tocathy):

Cathy, tell me a joke.

--------------------------------------------------------------------------------
cathy(tojoe):

Sure, here's one for you:

Why don't scientists trust atoms?

Because they make up everything!

--------------------------------------------------------------------------------
joe(tocathy):

Haha, that's a good one, Cathy! Okay, my turn.

Why don't we ever tell secrets on a farm?

Because the potatoes have eyes, the corn has ears, and the beans stalk.

--------------------------------------------------------------------------------
cathy(tojoe):

Haha, that's a great one! A farm is definitely not the place for secrets. Okay, my turn again.

Why couldn't the bicycle stand up by itself?

Because it was two-tired!

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

The comedians are bouncing off each other!

Summary​

In this chapter, we introduced the concept of agents, roles andconversations in AutoGen. For simplicity, we only used LLMs and createdfully autonomous agents (human_input_mode was set toNEVER). In thenext chapter, we will show how you can control when toterminate aconversation between autonomous agents.