Multi-Agent AI Systems: Production AI Agents with AutoGen

Set up your development environment with Python and VSCode, create and securely copy your OpenAI API key, and review the $5 credit and billing basics.

Learn cross-platform Python installation with a knowledge base that shows exactly how to set up Python on Windows, Mac, or Linux, and get you started.

Explore what an AI agent is, how it pairs with large language models to autonomously perform tasks, and how tools, APIs, and memory enable personalized, goal-driven decisions.

Autogen is an open source framework that lets developers build AI agents that communicate and cooperate. It orchestrates multi-agent conversations, human-in-the-loop workflows, tools, and pre-built workflows to improve LLM performance.

Install autogen in a virtual environment, set up an .env with your OpenAI API key, and build a simple Conversable agent using GPT-3 turbo with caching for building multi-agent systems.

Explore Autogen building blocks: multi-agent conversations and conversation patterns. See how LM agents, tools, and human input chat to coordinate tasks autonomously.

Configure a two-agent chat in AutoGen by instantiating an assistant agent and a user proxy agent, initializing the conversation, passing a system message, and tuning code execution and docker options.

Explore the multi-agent conversation framework flow in AutoGen, where a user proxy and an assistant agent execute tasks, handle errors, install packages, and generate outputs like charts and code.

AutoGen enables agents to execute code using built-in executors, including command line and Jupiter environments. Code runs locally or in Docker containers for isolation, returning outputs as messages.

Run a hands-on simple code executor with a two-agent multi-agent system to fetch Meta and Tesla stock data and plot a line chart.

Add human input to influence multi-agent plots by feeding feedback during execution, guiding agents to fetch and visualize historical stock prices for Google, Amazon, Microsoft, and Apple.

Explain the user proxy and assistant agent roles in AutoGen's convertible and Conversable hierarchies, and how default human-input modes—always for proxies, never for assistants—determine code execution versus writing.

Customize agents with a system message to provide context, like AWS, biology, astronomy, or history. Avoid subclassing by using a proxy agent and assistant agents.

Understand how human in the loop manages message flow between agents, using interception and optional human or auto replies to drive appropriate outcomes.

Mastering Autogen applies hands-on learning to building blocks, features, and agents—user, proxy, agent, and assistant—within a multi-conversation framework with code executors.

Pablo invites you to leave a review and share how the course adds value. Post questions in the discussion board to get answers and support peers.

mastering autogen explains three human input modes—never, terminate (default), and always—plus how skipping yields an auto reply, guiding how agents interact in the loop.

Explore never human input mode in a hands-on multi-agent setup where two agents play a silent guessing game about an animal, ending when the elephant is identified.

Explore hands-on always human input mode by using a human proxy to drive conversations without a language model, with hints and human-driven validation.

Demonstrate terminate mode by enforcing human input until termination and limiting auto replies to one. Show a guest animal guessing game where the elephant is identified through questions.

Learn how large language model caching reduces costs and improves reproducibility by using the AutoJane api request cache, the cache folder and database, and configurable seeds, max tokens, and settings.

Explore how AutoGen uses predefined tools to limit agent-written code. These tools support web searches, calculations, reading files, and external api calls through controlled availability.

Learn to build simple auto gen tools, add numbers and multiply numbers, wire two agents to call these tools, register tool signatures, and compute sums.

Build a real-world travel assistant by creating tools for flight status, hotel information, and travel advice, registering them with an agent to automate itinerary planning.

Develop and run a travel planner with multiple tools that calculate travel time, convert currency, and suggest activities for destinations like Paris and Tokyo, all managed by travel agents.

Explore autogen and tools, learn what tools are, how to create them, and how to use them with agents to complete tasks. Envision new use cases for your company.

Explore two-agent chats, sequential carry over, group and nested patterns, and learn how next-speaker strategies (round robin, random, manual) plus chat history and summarization shape multi-agent workflows.

Build a two-agent conversation with a travel planner and guide agent using the initiate_chat method, producing a LM-based summary from their chat history. Analyze message exchange, summaries, and caching costs.

Explore sequential chats in autogen as linked two-agent conversations that pass a carry over summary to the next agents, preserving context across subtasks, ideal for complex tasks.

Master the flow of sequential chat with multiple agents, including an initial, uppercase, word-count, reverse text, and summarization agents, carrying over context to produce a final summary.

Explore how the group chat manager coordinates multiple agents, selects a speaker, broadcasts messages, and repeats steps, with strategies and examples using arithmetic operator agents to reach a target.

Learn to build and orchestrate a multi-agent group chat using AutoGen by creating flight, hotel, activity, restaurant, and weather agents and coordinating them through a group chat manager.

Explore adding group chat to a sequential chat workflow, using a group chat manager and introductions to coordinate travel planning agents (flight, hotel, weather, restaurant, activity) and compute total costs.

Explore nested chat in origin, using a nested chats handler to encapsulate complex workflows within a single conversational interface for multi-agent systems, preserving history and context.

Explore nested chats with a writer, user proxy, and critic to generate and refine content through reflection and critique in a multi-agent workflow for product reviews.

Learn conversation patterns—two-agent chat, sequential chat, group chat, and nested chat—and how to compose blocks into complex workflows for diverse applications in multi-agent systems.

Explore building a customer service automation workflow with Autogen multi-agent flows, routing inquiries through classification, automated responses, knowledge base search, troubleshooting, feedback collection, and escalation to human support.

Automate monthly financial report generation and review by orchestrating data aggregation, detailed report creation, accuracy and compliance checks, and executive summary generation with feedback.

Automate research paper analysis and visualization by orchestrating multi-agent workflows that find relevant papers, analyze health care ML applications, and generate and save a bar chart of findings.