Your first model call on Azure

Already know a deployment name is not a model id? Skip to the next lesson.

Welcome to the Azure half. Your loop does not change. What changes is where it sends its request, and one naming rule that catches almost everybody.

Some words you will keep meeting

• Azure is Microsoft's cloud. Rented computers you never see.
• Azure AI Foundry is the part of Azure for building AI applications. It is where your models and agents live.
• A project is your workspace inside Foundry. Agents belong to a project.
• A deployment is one specific model, switched on and given a name by you.

That last one is the trap.

The trap

gpt-4o is a model id, the name of the model in Microsoft's catalogue.

When you switch that model on in your project, you choose a name for it. You might call it agents-chat. That is the deployment name.

Your code must send the deployment name. Send the model id and Azure rejects the request, often with an error that reads as though the model does not exist. It does exist. You asked for it by the wrong name.

People lose hours to this. That is why it is the first Azure lesson.

What to write

1. Set AZURE_DEPLOYMENT to agents-chat, not gpt-4o
2. Call chat(messages, deployment=AZURE_DEPLOYMENT) in your loop
3. Stop when there are no tool calls, exactly as before

This runs on recorded answers, so nothing costs money. The mistake is real, and the error you get is the real one.

The point

Same code, different address. The loop did not change at all.

Break it on purpose

Leave AZURE_DEPLOYMENT as gpt-4o and run. Read the error carefully. You will see it again in your own work, and now you will know what it means.