Use Azure OpenAI

Azure OpenAI serves models through deployments that you create in your own Azure resource. You add it in Settings, Administration, Model Providers like any other provider (see Adding a model provider): select Add a provider and choose the type Azure OpenAI. You need the admin ability.

Fields#

Field	What to enter
Endpoint	The resource's address, for example https://my-resource.openai.azure.com. It must start with https://, have no path, username or password, and end in an Azure domain: openai.azure.com, cognitiveservices.azure.com or services.ai.azure.com (or the same under azure.us or azure.cn). Addresses such as an IP number, localhost or another company's domain are refused.
API version	The Azure OpenAI API version every request uses, for example 2024-10-21.
API key	One of the resource's keys. It is encrypted and never shown again.
Deployments	One row per deployment Coworkers may use: the Deployment name, and optionally the Base model. Select Add deployment for more rows.

Add provider stays unavailable until the endpoint, API version, key and at least one deployment name are filled in.

Why deployments, and what the base model is for#

Azure routes a request by the name of your deployment, not by the model's own name, and gives the product no way to list your deployments. So the models offered for this provider are exactly the deployment names you enter, and the product sends each request to that deployment.

The deployment name can be anything you chose in Azure, so the product cannot tell which model it serves. The Base model (for example gpt-4o) tells it. The product uses the base model to look up the price of each call and the model's limits and abilities, such as reading images. If you leave it blank, the deployment name is used for the lookup, which works when the deployment is named after its model. A deployment whose model cannot be identified is shown as Unpriced in Usage & cost until you price it (see Price models the price list cannot price).

An embedding deployment is recognised by "embed" in the name or by its base model, and is offered as an embedding model.

Test#

Test checks that the endpoint is reachable and that Azure accepts the key. A rejected key reads "Azure rejected the key". It sends no model request. Whether a deployment exists is confirmed on the first call that uses it.

Edit#

Select Edit, change the endpoint, API version or deployments, and select Save changes. Leave API key blank to keep the stored key, except when you change the endpoint: then enter the key again, because the stored key is never sent to a different address on your behalf. The deployment list is replaced by whatever the form shows.

The stored endpoint, API version and key are the only ones used. Azure environment variables set on the server, such as AZURE_API_BASE or AZURE_API_KEY, do not change where a call goes or which key it uses.