Recipe for Instant Homemade AI Agents With Azure OpenAI GPT-4o Using .NET

This is a step-by-step tutorial for using Azure OpenAI Assistants for making a quick and delicious agent capable of RAG (Retrieval Augmented Generation) that you can interact with via simple .NET code. If you prefer watching video content just skip to the end of this post!

Start with a hub and a project

First you want to be in your "All hubs" in Azure AI Studio before you get started.

Click on the name of your Hub to activate it. Look at the tippy top and you know that you're in All hubs and your particular Hub has been activated. Next, click on "All projects"

Look at the tippy top again and you can see you're in the subroute of "All hubs > your hub > All projects." Go ahead and make a "New project." If you don't, the system won't work in an optimal way so just come along with me on this journey ;-).

You can go ahead and create a project. This can take roughly a minute.

Okay, you're in good shape now because:

1. You have a Hub
2. You have a Project
3. You clearly see you're using the Hub and the Project at the tippy top

Time to make an Azure OpenAI Assistant in the UI

Now let's go use the Azure OpenAI Assistants API route in Azure AI Studio. It's going to be cool!

You should see the Assistants Playground come up. And be sure to follow this sequence:

1. Click on "New assistant" even if you think you don't have to. Because ... you do!
2. For the Deployment field, if you don't have an existing gpt-4o model available, be sure to click on the tiny blue text that reads "Create new deployment" to create an instance. If you're in a region where gpt-4o isn't available, then this isn't going to work. Follow these quick instructions to set up a model in Azure AI if this is all new to you.

LLMs will by themselves "make stuff up" unless you ground them in information that matters to you. There's an easy way to do a so-called "RAG" (Retrieval Augmented Generation) by clicking on the little "Tools" tab and adding files to the "File search" feature.

And then a wickedly complex screen pops up like the following. Ignore most of what you see and pay attention to the lower left button that reads "Upload local files" and select a file that's supported to upload it. And then after your file has uploaded, be sure to press "Attach" so that the file gets associated with your assistant named "Assistant671."

You should see the good news after hitting "Attach" like so:

Assuming you have turned on "File search" you can ask any question about the document you've uploaded. I uploaded a fictional travel itinerary and am asking Assistant671 to use the file I've uploaded as the grounding for a response.

Pretty cool, huh? Now let's do that same thing in code instead. It'll feel a lot easier :-).

Creating an Azure OpenAI Assistant API agent with file upload in code

Visit this easy-peasy repo to get going in .NET by "edandersen" on GitHub.

Clone the repo and all you need do is add your Azure OpenAI "endpoint" and "key" in the Program.cs file:

Where do you get that info in Azure AI Studio? Well, do you remember the "Assistants playground" where you were trying out Assistants API? Click on the "</> View Code" button.

And you'll see some code you could include instead of the GitHub repo I've recommended above. But for now, just grab the Endpoint and API Key.

With these two pieces of information added to the Program.cs file, when you do a "dotnet run <filename>" you will be able to upload a file easily.

And the right things happen. Yaaaay!

Take the next step

Are you intrigued? Then I suggest you take the next step and try out multi-agent systems that run on top of Azure OpenAI Assistants API as demonstrated by AI Chef Marco Casalaina on the Cozy AI Kitchen show:

Or watch how agents are reshaping the future of UX with Kwame Nyanning, the world's expert on "Agentic UX":