Bring your AI key

The workshop calls two AI models:

GPT-4.1 for chat: summaries, show notes, tool calls and the agent
GPT-4o Transcribe for speech-to-text: turning the episode audio into a transcript

You'll need your own key from one of two providers:

Path	Choose it if…	What you'll need
OpenAI	You have, or can create, an OpenAI API account	An API key and a few dollars of prepaid credit
Microsoft Foundry	Your company already uses Azure	An Azure subscription where you can create resources

Whichever path you pick, you end up with three values: an API key, a chat endpoint and a transcription endpoint. In Lesson 0, Step 8 you store them as .NET user secrets, and every lesson branch picks them up from there.

Note

Adding billing (OpenAI) or deploying models (Foundry) can take a while, so set up your key before you start Lesson 0.

Tip

What will it cost? Very little. Transcribing the four-minute episode costs a few cents each time, and a summary, the show notes or a full agent run costs a few cents more. Expect well under US$1 for the whole afternoon.

Set up your provider

Pick the tab for your path.

OpenAI
Microsoft Foundry

About 10 minutes.

Sign in at platform.openai.com, or create an account there. A ChatGPT subscription doesn't include API access. The API is billed separately.
Add credit. Open Settings → Billing and add a payment method, then buy some credit. The minimum top-up (US$5 at the time of writing) is far more than the workshop needs. Without credit, every call fails with "You exceeded your current quota".
Create a project (recommended). Use the project picker at the top left to create a project called Umbraco Workshop. It keeps the workshop's usage separate and lets you revoke everything in one place afterwards. In the project's settings you can also set a budget.
Create an API key. Go to API keys and choose Create new secret key. Pick your Umbraco Workshop project, leave Permissions on All, and create it. Copy the key straight away, because OpenAI only shows it once. Keep it in a password manager.
Check it works. Open the Playground in the same project, choose the gpt-4.1 model and send "Hello". A reply means your billing and model access are sorted.

Your three values:

Value	What to use
API key	Your new key. It starts with `sk-`.
Chat endpoint	`https://api.openai.com/v1/`
Transcription endpoint	`https://api.openai.com/v1/` (the same one)

Note

New OpenAI accounts start on the lowest usage tier. Its rate limits are fine for one person working through the workshop.

About 15–20 minutes. You need an Azure subscription where you're allowed to create resources. Some free-trial and sponsored subscriptions can't deploy these models. If yours can't, use the OpenAI path.

Create a Foundry project. Sign in to the Microsoft Foundry portal and create a new project. This also creates a Foundry resource to hold it. Choose the East US 2 region, which has both models. Make a note of the resource name you choose.
Deploy GPT-4.1. Open the model catalog, find gpt-4.1 and deploy it. Choose the Global Standard deployment type and keep the deployment name as gpt-4.1. If the portal says you're out of quota, lower the tokens-per-minute limit (50K is plenty) and try again.
Deploy GPT-4o Transcribe. Do the same for gpt-4o-transcribe, and keep the deployment name gpt-4o-transcribe.
Copy the key. The project's Overview page shows an API key and your endpoints. Copy the key. The endpoints start with https://<your-resource>.services.ai.azure.com/, where <your-resource> is the resource name from step 1.
Check it works. Open the gpt-4.1 deployment in the playground and send "Hello".

Your three values (replace <your-resource>):

Value	What to use
API key	The key from the Overview page
Chat endpoint	`https://<your-resource>.services.ai.azure.com/openai/v1/`
Transcription endpoint	`https://<your-resource>.services.ai.azure.com/openai/deployments/gpt-4o-transcribe?api-version=2025-03-01-preview`

Important

Keep the default deployment names. When Umbraco asks for the model gpt-4.1, Foundry looks for a deployment with that name. If you named yours differently, calls fail with "deployment not found".

Why does Foundry need two endpoints?

Foundry's OpenAI-compatible /openai/v1/ endpoint serves chat, but it doesn't serve audio transcription. We tested this in September 2026. So transcription calls go straight to the transcription deployment's own URL instead. You'll see how Umbraco AI absorbs that difference in Lesson 1, and it's a nice example of why connections exist.

Using a classic Azure OpenAI resource instead of a Foundry resource? That works too. Use https://<your-resource>.openai.azure.com as the start of both endpoints.

Next: save them as user secrets

Head to Lesson 0, Step 8 and pick the tab for your path. User secrets live in your user profile, not in the repository, so they can't be committed by accident, and they stay put when you switch branches.

Warning

Treat the key like a password. Don't paste it into chat, a screenshot or a commit. When the workshop's over, revoke it in your provider's portal (or delete the Umbraco Workshop project) unless you want to keep experimenting.

Table of Contents

Bring your AI key

Note

Tip

Set up your provider

Note

Next: save them as user secrets

Warning