Get an Azure OpenAI key

Who this is for: anyone whose company runs OpenAI models in Microsoft Azure and wants SpiderGate to use that resource.

What this does: walks you through finding your Azure OpenAI key and endpoint, deploying a model, and adding the resource to your SpiderGate vault.

Before you begin

An Azure subscription, and permission "to create Azure OpenAI resources and to deploy models". Microsoft says most models need no application form.
About 15 minutes.

Limits and terms to know first

Azure calls models by deployment name, not model name: "Azure OpenAI always requires deployment name, even when using the model parameter." You must deploy a model before you can call it. Your Azure agreement and quota apply.

Steps

Step 1 — Create the resource

In the Azure portal, select Create a resource, search for Azure OpenAI, and select Create. Fill in Subscription, Resource group, Region, Name and Pricing Tier (Standard), then create it and select Go to resource.

Step 2 — Deploy a model

In Foundry (classic), open the resource, select Deployments under Shared resources, then + Deploy model → Deploy base model. Note the deployment name you give it.

Step 3 — Copy the key and endpoint

In the resource, open Resource Management → Keys & Endpoint. Copy the endpoint (for example https://your-resource.openai.azure.com/) and either KEY1 or KEY2. Two keys let you rotate one without downtime.

Step 4 — Add it in SpiderGate

Open Gate → Vault (/dashboard/gate/vault) and click Add Provider. Pick Azure OpenAI from the Enterprise group, then fill in:

API Key — KEY1 or KEY2.
Endpoint URL — the endpoint from Step 3.
API Version — optional. Microsoft's latest dated GA version is 2024-10-21; the 2024-02-01 in the field is only a placeholder. If you leave it empty, SpiderGate's check uses 2024-10-21.
Label — a name you will recognise.

Under Routing, turn Share with SpiderIQ key pool off, so the key stays with your brand. Click Add key.

Verify it worked

When you click Add key, SpiderGate sends one request to Azure with the key before it stores anything: it lists the models on your endpoint, https://<resource>.openai.azure.com/openai/models, with the key in the api-key header.

If Azure accepts the key, the dialog closes and the key appears in your vault list.

If it does not, the dialog stays open and shows:

The azure_openai key did not pass its check: <reason>. Nothing was saved.

The reason is SpiderGate's own wording, never Azure's error text. The usual ones are Invalid API key (401 Unauthorized), Access denied (403 Forbidden) and The account behind this key has no credit left (402).

To see the key serve a request, give it a slot in one of your own routing aliases and send a request to that alias. The trace shows which provider answered.

When you give the key a slot in a routing alias, use your deployment name as the model.

Troubleshoot

The Azure endpoint must look like https://<resource>.openai.azure.com. SpiderGate accepts only an https:// endpoint on an Azure OpenAI host (.openai.azure.com, .cognitiveservices.azure.com or .services.ai.azure.com). Copy it again from Keys & Endpoint.

The Azure API version should look like 2024-10-21. The API Version must be a date like 2024-10-21. Clear the field to use the default.

A request names a model that does not exist. Azure needs the deployment name from Step 2, not the model's catalogue name.

Too many keys checked for this workspace; try again in …s. SpiderGate runs at most 20 key checks per brand in any 10 minutes. Wait the number of seconds it gives, then click Add key again.

Request timed out (15s) or Could not reach … to check the key. Try again later. The provider did not answer in time. Nothing was saved. Try again in a minute.

The provider is rate-limiting us (429). Try again in a minute. The provider throttled the check itself. Wait a minute and retry.