Calling Azure OpenAI from Power Automate

AI Builder prompts are the easiest route to GPT in Power Platform. Sometimes, though, you need your own Azure OpenAI deployment — to pick a specific model, keep data in your own Azure region, or share one deployment across apps. Here's how to call it from a flow.

AI Builder or Azure OpenAI?

What you need from Azure

From your Azure OpenAI resource, note three values: the endpoint (for example https://my-resource.openai.azure.com), the deployment name you gave the model, and an API key. You also need the API version your resource supports — check the Azure OpenAI documentation for the current one.

Keep the key out of the flow

Never paste the key into the HTTP action. Two good options:

• Environment variable of type Secret, backed by Azure Key Vault — the cleanest choice for solutions that move between environments.
• At minimum, a Secret environment variable so the key isn't visible in the flow definition.

Build the request

Add an HTTP action:

• Method: POST
• URI: https://my-resource.openai.azure.com/openai/deployments/{deployment}/chat/completions?api-version={version}
• Headers: Content-Type: application/json and api-key: <your secret variable>
• Body:

{
  "messages": [
    { "role": "system", "content": "You summarise Dataverse case notes for managers in 3 bullet points." },
    { "role": "user", "content": "@{triggerBody()?['notes']}" }
  ],
  "temperature": 0.2,
  "max_tokens": 300
}

Read the answer

The model's reply is inside the first choice. Use this expression in a Compose action:

body('HTTP')?['choices']?[0]?['message']?['content']

Also log body('HTTP')?['usage']?['total_tokens'] so you can track cost per run.

Handle failures

Wrap the HTTP call in a Try scope. In the Catch scope, check the status code:

• 429 means you hit a rate limit — retry later with a delay.
• 400 often means the content was filtered or the body is malformed.
• 401 means the key is wrong or expired.