How do I connect Azure OpenAI?

Where: Settings → Content & publishing → Your AI accounts

Open this page: admin.<your-site>/settings/ai/providers (replace <your-site> with your site address)

Before you start

• An Azure account with an Azure OpenAI resource.
• At least one model deployed in that resource. Azure runs a model only after you deploy it.
• You are allowed to read the resource's keys in Azure.
• Only members who manage AI can connect, and you need to have signed in recently.

Steps

1. Open the Azure portal and open your Azure OpenAI resource.
2. Go to Resource Management, then Keys and Endpoint.
3. Copy the Endpoint. It looks like https://my-resource.openai.azure.com. It must end in .openai.azure.com or .cognitiveservices.azure.com.
4. Copy KEY 1. Keep KEY 2 as a spare. Your admin accepts a key of 32 characters (letters a–f and digits) or 84 characters.
5. Find your deployment names. In the Azure AI Foundry portal, open your resource's Deployments. The deployment name is the name you gave when you deployed the model. It can differ from the model's own name, for example gpt-4o-mini.
6. In your site admin, open Settings → Content & publishing → Your AI accounts, and find the Azure OpenAI card.
7. Paste the Azure endpoint. Add each deployment under Deployment names, and tap Add after each. Add at least one.
8. Paste the key and tap Test & connect. You see "Checking your key…" and then Connected ✓.

Good to know

• You can connect up to two Azure resources.
• Azure has no model list for a key, so the admin only shows the deployments you typed. Add more later if you deploy more models.
• Prices are optional. With prices, the admin can show costs. Without them, the AI does not switch to this account on its own when another account fails.
• Azure bills you. The Azure portal has your costs and budgets. AI budget and limits
• The key stays on your own server. Keynate never sees it.

Common problems

Use your Azure OpenAI endpoint

The endpoint must end in .openai.azure.com or .cognitiveservices.azure.com. Copy it from Keys and Endpoint. A custom domain or an IP address does not work.

This model or deployment isn't available on your Azure OpenAI account

The deployment name does not match. Names must match exactly, including capitals. Check Deployments in Foundry, fix the name on the card and test again.

Azure OpenAI refused this key

Copy the whole key again. Keys belong to one resource, so use a key from the same resource as the endpoint.

We couldn't reach this address

Check the endpoint for typing mistakes. If your resource only allows certain networks, the test can fail. Allow public access to the resource, or add your server's address, then test again.

Azure says the quota or rate limit is reached

Wait a minute, or raise the quota of the deployment in Azure.

Questions

Related

• Use any AI with your own key
• AI budget and limits
• AI not working