AI in C# using the Microsoft Agent Framework

Having taken a similar course like this myself, I know you want to see some code now!... So we will leave the theory behind a bit, and see Microsoft Agent Framework in action, well knowing that we are skipping a few steps here... Don't worry; we will go into more details after Hello is done.

We will, in this lecture, install the needed NuGet package and make the raw connection

We will use the raw connection and make our first Agent and call to the LLM

To see our options, let's replace OpenAI as the provider with Azure OpenAI as the provider instead (to show that you only switch the raw connection, and the rest of the code will remain the same)

Let's see how to properly store secrets for LLMs in C# code using User-Secrets + Go to OpenAI / Azure OpenAI and revoke the exposed secrets.

In this Optional video, I show that it is not only OpenAI and Azure OpenAI that the Microsoft Agent Framework can't connect to, but literally anything is supported! You can skip this if you only intend to use OpenAI

Note (July 2026): GitHub Models are being retired so they can't be used anymore.

In this lecture, I will give you a brief introduction to the Companion GitHub Repository

Before we go into the main content of the course, I want to touch on the subject of cost to follow along. It is a rather lengthy but important topic... If you do not care about cost, you can skip it for now and see it at the end for reference.

It is time to introduce the 4 main concepts of AI, and what the main part of the rest of this course is about

In this video, we talk about why "Chat" is not really a covering word anymore, and how we use "Chat" for not just chatbots.

In this first sample, we dive a little deeper into what we actually did during Hello World, and introduce the option to do streaming calls

In this lecture, we will introduce the Chat-loop and how to make the AI able to do follow-up questions using an AgentSession.

In this lecture, I will give an introduction to Prompt engineering, in the form of setting Instructions on the Agent

In this optinal lecture, we will explore all the options there are for setting options when creating agents.

In this optional lecture, we will explore the options you have when making the call to the agent beyond the message and AgentThread we have already seen.

In this lecture, we are going to define, create, and use our first AI Tools

In this lecture, we will learn how to consume an MCP Server

In this lecture, we will introduce Tool Calling Middleware that lets you log, change, and refuse tool calls.

In this lecture, I will show you that you can use Agents as Tools for other Agents

In this lecture, we explore how we can feed up-to-date information into our agent when the user asks for current news items, beyond AI's cut-off date.

In this lecture, I will teach you about the code_interpreter tool, that let an AI execute python script in a sandbox container online, for complex operations or example chart-generation.

In this lecture, we will finally look into Structured Output: what it is and the problem it solves

Let's explore how Structured Output can have Instructions on the object

As we are on 3/4 of the way through the 4 main concepts of AI, we now have enough knowledge to see what actually happens when we call an LLM in terms of requests to the LLM and responses back

In this optional video, I will show you how to enable the option to see the raw Request/Response of an LLM

Welcome to the RAG section of this course; RAG is a complex topic with many moving parts, so in this first lecture, we will talk about what RAG actually is before we get into the code

In this lecture, we will take what we learned in the previous one and make it more abstract in what parts of tech are involved and what the flow is to making RAG happen.

In this Lecture we will see how Embedding Vectors are generated and check how similar they are to each other

Now that we know about Vectors, let's choose a vector store to ingest the vectors into.

Now that we have our Vector Store, let's use it in a sample Chatbot and see the result

In this lecture, I will show a variation on how to use RAG as a Tool instead of always searching the Vector store upfront.

You made it through the 4 main concepts of AI and IMHO are able to make at least 90% of all the AI Apps you need to make. This lecture introduces the rest of this course.

In this video, I explore the Agent Framework HarnessAgent that allows you to build tools like Codex and GitHub Copilot via the Agent Framework