Connect Azure OpenAI to Xcellerate AIG

Connecting Azure OpenAI to Xcellerate AIG sends AI traffic to your own Azure OpenAI resource: your subscription, your region, your deployments. Your applications keep calling AIG's OpenAI-compatible API with a virtual key, and AIG turns each request into Azure's format. This guide is for admins who set up AIG for an MSP, SME or service business.

TL;DR

• You need your resource endpoint, an API key and the names of your deployments.
• Callers address a deployment as <provider-name>/<deployment-name>.
• One provider can front several Azure resources or API versions, one per key.

Before you start

• An Azure OpenAI resource with at least one model deployment. Write down each deployment name exactly as you named it in the Azure portal.
• The resource endpoint, for example https://<your-resource>.openai.azure.com, and one of its API keys. Both are on the resource's Keys and Endpoint page.
• Optionally, an API version. If you leave it empty, AIG uses 2024-10-21. That is a minimum, not a recommendation.
• The Admin role in AIG, or a custom role with permission to create and edit providers.
• Outbound HTTPS from AIG to your endpoint. No callback URLs are needed.

Set up Azure OpenAI

1. Create the provider

Go to Connect → Providers and click Connect a provider (or Add provider). Choose the provider type Azure OpenAI and give it a name, for example "Azure West Europe". Azure has no default Base URL: the endpoint is set on the key. A provider Base URL only serves as a fallback.

2. Paste the API key

Paste the Azure key into API key.

3. Fill in the Azure fields

These fields show their technical names in every language:

• endpoint (required): your resource URL.
• api_version (optional): the API version.
• deployment (optional): pins this key to one deployment.

4. Test and enable

Click Test connection, then Enable provider.

5. Call a deployment

Callers use <provider-name>/<deployment-name>. For example, azure_openai/prod-gpt4o reaches the deployment prod-gpt4o. The model part of the address is the deployment name.

6. Set your prices

Under Cost → Pricing, add a pricing override. A private deployment name may not be in the public datasheet, and negotiated prices differ.

What happens after you connect

AIG sends each request to <endpoint>/openai/deployments/<deployment>/<operation>?api-version=<version>, with your key in the api-key header. Chat, embeddings, images, speech and transcription are available. If your resource has fewer deployments, narrow the provider with What this provider serves; other requests are then refused inside AIG. Governance, logging, budgets and fallbacks apply as for any other provider.

Good to know

• Azure needs the deployment name, not the model name. AIG sends the model part of the address as the deployment, unless you set deployment on the key.
• Two subscriptions on different API versions? Add two keys, each with its own endpoint and api_version.
• AIG authenticates to Azure OpenAI with an API key.
• Your data is processed in the region of your Azure resource.
• Keys are write-only and never sent back to the browser.
• Without a licence, AIG runs in free mode: 5 virtual keys with 5 requests per key per day.

Troubleshooting

• Azure provider '<name>' has no endpoint configured on its key. Fill in endpoint on the key, or a Base URL on the provider.
• <model> did not answer (<ms> ms): <message> during the test. Usually the endpoint, deployment name or API version is wrong.
• No healthy API key is available for provider '…' Azure rejected the key and AIG marked it Dead. Fix the key, then click Clear health flags.

Talk to us about AIG

Want to know how AIG brings your Azure setup and your other providers together? Talk to us about AIG. Read more about every provider in one catalogue and virtual keys.