Azure OpenAI in Flow (Premium)

Azure OpenAI connector in Power Automate: prerequisites, Premium licensing requirement, and why Azure costs run separately from the license.

If you already run your own Azure OpenAI resource, you don't need to switch to AI Builder to use it in Power Automate. Microsoft provides a dedicated connector that links your flows directly to your Azure OpenAI deployment, including chat completions, embeddings, and optional integration with your own data via Azure AI Search. However, the connector belongs to the Premium class, which has consequences both for your Power Automate license and for your ongoing Azure costs.

This article shows what the Azure OpenAI connector delivers in Power Automate, which prerequisites you need to set up before your first flow, how the Premium licensing requirement plays out in practice, and why the Azure bill runs separately from the Power Automate license. It is based on Microsoft's official connector reference and licensing documentation.

What the Azure OpenAI connector delivers in Power Automate

The Azure OpenAI connector is a Microsoft-native connector and, according to the connector reference, is classified as Premium in Power Automate and Power Apps, also Premium in Copilot Studio, and Standard in Logic Apps. It is available in practically all Power Automate regions, except US Government (GCC), US Government (GCC High), the China Cloud operated by 21Vianet, and the US Department of Defense (DoD).

According to the documentation, the connector offers four actions:

  • Creates a completion for the chat message: classic chat completion against a deployed model, with parameters such as temperature, top-p, max tokens, presence penalty, and frequency penalty.
  • Get an embedding: vector representation for a single input text, for example for similarity search or classification.
  • Get multiple embeddings: the same function for an array of input texts in a single call.
  • Using extensions to create a completion for chat messages: chat completion with extensions that let you bring in your own data from Azure AI Search as context, the so-called "On Your Data" scenario.

In practice, this means a flow can send incoming text, for example from a form or an email, directly to your own Azure OpenAI deployment and then process the response in the next step, without a detour through a third-party connector.

Prerequisites before your first flow runs

Before the connector can be used, several things need to be in place according to the connector reference:

  • An active Azure subscription.
  • Access to Azure OpenAI itself; the service requires registration and is only available to approved enterprise customers and partners.
  • An Azure OpenAI resource created in a supported region with a supported model, for which at least the Cognitive Services Contributor role must be assigned on the resource.
  • Optional, for the On Your Data scenarios: an Azure AI Search resource with appropriately prepared data.

This groundwork happens entirely in the Azure portal and has nothing to do with Power Automate itself yet. Only once the resource is in place can you set up a connection to it in the flow.

Setting up the connection: API key or Managed Identity

According to the connector reference, the Azure OpenAI connector supports three authentication types:

  • API key: you enter the name of the Azure OpenAI resource and the API key, optionally along with the endpoint URL and key for Azure AI Search. This connection can be used in all regions and can be shared with other users.
  • Managed Identity: sign-in via a managed identity, according to the documentation currently only available for Logic Apps, also shareable.
  • Standard (deprecated): the old connection type without an explicit authentication type, present only for compatibility reasons and not shareable, everyone has to create their own connection when sharing.

For new flows, the API key is usually the most practical option, because it works in all regions and can be shared as a connection within the team, without every colleague having to create their own.

Premium licensing requirement: who on the team needs what

Because the connector is classified as Premium, the regular Power Automate licensing for Premium connectors applies. According to the Power Automate license types, entitlement for Premium connectors is included in the Power Automate Premium user license, the Power Automate Process license, and the Power Automate Hosted Process license. The free Power Automate license, by contrast, only covers Standard connectors, so a flow with the Azure OpenAI connector will not run on it.

How this plays out in practice depends, according to the Power Automate licensing FAQ, on the flow type:

  • Automated and scheduled flows run in the license context of the owner, regardless of who triggers the flow or which connections it uses. If such a flow uses the Azure OpenAI connector, only the owner needs a Premium license.
  • Instant flows, meaning flows started by a button or a manual trigger, run in the context of the person who starts them. If the flow contains the Azure OpenAI connector, every person who starts it manually needs their own Premium license.
  • Instead of licensing every individual person, a flow can instead be equipped with a Process license. It then runs regardless of whether individual users have a Premium license, which is especially worthwhile for flows started by many people on the team.

For a team that wants to use the Azure OpenAI connector in a shared, manually started flow, this means in practice: either every person involved gets a Premium license, or the flow itself is equipped with a Process license. This way you stay in control of who is actually allowed to use the connector, instead of sorting out the licensing question only after the first error in the flow.

Azure costs come on top of the Power Automate license

The Premium license only covers access to the connector within Power Automate. Actual use of the language model runs through your Azure OpenAI resource and is billed separately through your Azure subscription, based on consumption for the respective model deployment. These costs arise regardless of whether the call comes from Power Automate, from a Logic App, or from your own application, because billing happens at the level of the Azure resource, not at the level of the connector.

Two points from the connector reference are particularly relevant for planning:

  • Access to Azure OpenAI itself has to be requested and is not automatically available with every Azure subscription; you should plan for this early, before a project deadline depends on it.
  • The throttling limit is 1,000 API calls per connection per 60 seconds. For high-volume flows, such as bulk document processing, it's worth considering multiple connections or batching requests so you don't hit this limit.

Anyone who calculates both cost blocks, the Power Automate Premium license and Azure OpenAI consumption, separately from the start avoids unpleasant surprises on the first Azure bill after go-live.

Known limitations

The connector reference lists three limitations you should know before building a flow:

  • As a data source for the On Your Data feature, only Azure AI Search is supported; other vector databases cannot be connected through this connector.
  • Function calling is not supported by the connector.
  • Images in message content are not supported; the connector is designed for pure text interaction.

Anyone who absolutely needs one of these scenarios, for example function calling for structured returns, has to fall back on a custom connector against the Azure OpenAI REST API.

At NordFlux, we build Azure OpenAI integrations as part of our Power Automate consulting in a way that license and Azure costs are transparent from the start, instead of only becoming visible after the first production flow.

Frequently asked questions

Do I absolutely need a Premium license for the Azure OpenAI connector?

Yes. The connector is classified as Premium, so the regular Power Automate licensing for Premium connectors applies. The free license only covers Standard connectors and is not sufficient for the Azure OpenAI connector.

Is a single Premium license enough for the entire flow?

That depends on the flow type. For automated or scheduled flows, the owner's Premium license is enough. For manually started instant flows, on the other hand, every person who triggers it needs their own Premium license, unless the flow is instead equipped with a Process license.

Does the Power Automate Premium license also pay for Azure OpenAI usage?

No. The Premium license only pays for access to the connector in Power Automate. Actual model usage runs through your Azure OpenAI resource and is billed separately, based on consumption, through the Azure subscription.

Can I use my own company data as context with the connector?

Yes, using the extensions action for chat completions, you can bring in your own data material, but according to the connector reference only via a connected Azure AI Search resource as the data source. The connector currently does not support other vector databases.

What happens when a flow reaches the throttling limit?

The throttling limit is 1,000 API calls per connection within 60 seconds. If it is exceeded, the connection blocks further calls until the time window expires. For high volumes, it helps to batch requests or set up multiple connections for different flows.

About NordFlux

NordFlux UG (haftungsbeschränkt)

NordFlux builds digital employees for organisations: automations and AI agents that take over repetitive work. You stay in control.

More about us
Free initial analysis

Concrete questions about automation or AI?

In a free initial analysis we discuss your case directly. No strings attached.