Unanswered

Help Connecting Azure OpenAI Model with Data Source in Power Automate

(2) Share

Report

Posted on by H.Hassan

Hi everyone,
I am currently implementing a Power Automate flow that connects to an Azure OpenAI model.
Here's what I’ve done so far:

I created and deployed an Azure OpenAI model using Azure AI Foundry.
I used the text-embedding-ada-002 model for vector search. The vectorized data is stored in Azure SQL Database.
I integrated this data with Azure AI Search by importing and vectorizing it.
I then deployed a GPT-4o model on Azure AI Foundry, configured instructions, and set Azure AI Search (connected to the embedded data) as the data source.
When I test the model in the Azure AI Foundry chat interface, it works correctly. It follows the instructions and fetches relevant data from the database.

The issue: When I try to use this model in Power Automate via an HTTP action:
I use the endpoint and API key from Azure AI Foundry.
In the body of the HTTP request, I use the same JSON structure that works in the Foundry interface.
I use a Compose action to extract the response

However, the model's response in Power Automate does not fetch data from the database correctly. Sometimes it says "no data found", or returns irrelevant data — even though the same input gives correct results within Azure AI Foundry.

My questions are:
How can I properly connect the Azure OpenAI model along with its data source in Power Automate?
Do I need to make a separate HTTP call to the embedding model as well?
If yes, what should I include in the Body section of that request?
Or is there another recommended approach to achieve this integration?
Is there any way to change the output format of the response we get in the Azure AI Foundry chat?
For example, to return data in a custom JSON or a more structured format instead of plain text.

Thank you!

Categories:

Building flows

I have the same question (0)

All responses (7)

Answers (0)

Lagortinez 128 Super User 2025 Season 2 on at

Like (1)

Report

Hi there!

I see that in the content, you're using the outputs of the compose, and giving it to the API Params already. Since you stablished the Data Source in Foundry, I think you don't need to pass it again from Power Automate, that wouldn't make sense to me.

One piece of advice I would give you is that you play with the network tab, trying to isolate the API Call you are making in foundry, and mimic the same behaviour in Power Automate.

I fear this might be linked with the vector database behind it, but finiding the request behind it might help you out :)

Was this reply helpful? Yes No
stampcoin 5,058 Super User 2025 Season 2 on at

Like (1)

Report

@H.Hassan Hi,

The HTTP post body, is a place that you have to implicitly specify what you want the API to do.

There is no “magic” behind the scenes in Power Automate: whatever the service expects, you must put into your JSON payload.

The current one you got is kind of you have the access to the endpoint, that's all.

you can consider this as that you send a shopping list to someone, which only contains the destination.

so wrap the post body, so the other side knows exactly what to do.

good luck.

Was this reply helpful? Yes No
H.Hassan 14 on at

Like (0)

Report

@Lagortinez
Thanks for your response!

I removed the Compose action and tested again, but I’m still not getting a response from the added data source.
Regarding the Network tab suggestion — could you please explain a bit more?

Was this reply helpful? Yes No
H.Hassan 14 on at

Like (0)

Report

@stampcoin
Thanks for the reply!
Right now, I’ve added the system and user roles in the JSON body, along with parameters like temperature, as shown in the screenshot I’ve attached.
Could you please guide me on what exactly I need to include in the JSON body to enable vector search, so the model fetches data from the added data source?

Appreciate your help!

Was this reply helpful? Yes No
stampcoin 5,058 Super User 2025 Season 2 on at

Like (0)

Report
@H.Hassan Hi,

I think you need to implicitly specify the AI Search Endpoint in this scope, you data source.

Request body:

The doc tells that ( as what I understand):

Where ( search endpoint and index name).

How ( which embedding model).

Authentication/authorization ( search API key).

Below is python snippet from Microsoft, for HTTP post, it should be the same or similar.

Examples:

If you have more parameter, you wrap it in the body.

feel free if you have any questions.

Was this reply helpful? Yes No
H.Hassan 14 on at

Like (0)

Report

@stampcoin
should I need to add all of these?

Was this reply helpful? Yes No
stampcoin 5,058 Super User 2025 Season 2 on at

Like (0)

Report

@H.Hassan Hi,

Maybe you can specify the mandatory ones, once that works, you add more fields.

Use python or other tools will be much easier for testing than power automate, once you make progress, will be easy to take action in power automate.

good luck.

Was this reply helpful? Yes No