How to deploy Azure OpenAI chat with Copilot Studio Agent

Welcome to Microsoft Q&A and Thank you for reaching out.

Now you want to deploy the solution as a conversational agent inside Microsoft Copilot Studio, not as a standalone web app.

Create an Azure Function App or Web API in C#, Python.

Implement this logic:

Accept natural language input from the user.

Send it to your Azure OpenAI endpoint.

Use Azure AI Search to enrich context.

Generate and execute a dynamic SQL query on Azure SQL Database.

Return the final response as JSON.

Deploy it to Azure and note the public endpoint URL. 2: Prepare Knowledge Sources in Copilot Studio

Copilot Studio can directly connect to:

Azure AI Search

Custom APIs (your backend)

To add Azure AI Search: Go to copilotstudio.microsoft.com

Select your environment.

Click Knowledge → Add a source

Choose Azure AI Search

Enter your: Search service name, Index name,API key,Region

3: Create a New Agent in Copilot Studio

To create an agent:

Open Microsoft Teams and Install Copilot Studio from the Teams app store

Open Copilot Studio and Click Create a copilot and Name your agent and choose the correct team and environment

4: Connect Your API via a Custom Action

Now, connect your backend API to Copilot Studio.

To add your API:

In your Copilot project, go to Plugins → Custom actions

Click + Add a custom action

Select Create from API definition

Upload your OpenAPI (Swagger) spec OR manually define:

API name

Endpoint URL

HTTP method

Request/response schema

Headers

After saving, test the connection and action.

5: Build Your Chat Flow

Now set up how your Copilot handles input and returns dynamic SQL results.

To build the flow:Go to Topics → Create a new topic (“Dynamic SQL Query”)

Add trigger phrases (Like “How many sales last week?”, “Show me revenue by region”)

In the flow editor:

Add a message to acknowledge the query

Add a step to call your custom action (API)

Add a message to display the returned result to the user

6: Test Your Agent

Copilot Studio has a built-in test chat window.

To test:

Click Test your copilot and Enter a natural language question Confirm it:

Sends to your API

Generates and executes SQL correctly

Returns useful data to the user

Fix any issues with response format or API timeout.

7: Publish & Deploy to Microsoft Teams

To publish:

Click Publish → Next → Publish now

To deploy to Teams:

Go to Channels

Click Microsoft Teams

References:

learn.microsoft.com/en-us/microsoft-cloud/dev/copilot/overview#build-your-own-copilot

learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/responses?tabs=python-key

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thank you!