Bring your own model for your prompts

The Azure AI Foundry models for prompts feature is available within Copilot Studio, Power Apps, and Power Automate. This native integration with Azure AI Foundry brings the latest frontier models into your prompts in Copilot Studio.

The feature includes the following models:

OpenAI GPT 4.5
Llama
DeepSeek
And 1,800 plus more in Azure AI Foundry / Model catalog

You can access a diverse portfolio of AI models, including cutting-edge open-source solutions, industry-specific models, and task-based AI capabilities—all within the trusted, scalable, and enterprise-ready platform of Copilot Studio. With the native Azure AI Foundry integration, we bring the best capabilities from across our platforms into Copilot Studio to make it your one-stop hub to build agents.

Create an agent

To use this feature, create an agent in Copilot Studio for your scenario.

Sign in to Copilot Studio.
In Copilot Studio, create an agent for your scenario by selecting Agents > New agent > Skip to configure.
Fill in the fields, and then select Create.

Learn about alternative instructions in Create an agent.

Add a prompt in an agent

You can add a prompt directly as an action to an agent, or add a prompt to a topic.

Add a tool in an agent

In your agent, select the Tools tab, and then select Add a tool.
Select any of your existing prompts, or select New tool and select a prompt action.

Add a prompt as a topic

Alternatively, you can create a prompt directly in the Topics tab in Copilot Studio and add it to the agent.

In your agent, select the Topics tab.
Select the plus sign (+) to add a node where you want to add a topic.
Select Add a topic > New prompt.

Add and configure a prompt

After you create an agent, add prompt as an action to create a prompt.

In Copilot Studio, select New tool > Prompt.

The Tools tab contains the tools you can use in Copilot Studio. To use models from Azure AI Foundry, use the Prompt tool.
At the top of the screen, enter a name for your prompt.
Enter your prompt in one of the following ways:
- Use your own prompt: Under the Instructions heading, write or paste your own prompt.
- Get help from Copilot: Under the Get started with Copilot heading, describe what you want your prompt to do, and then select the right arrow. Copilot suggests a prompt for you. If you want to use it, select Keep it.
While you work with prompts, choose a model you want to work with. By default, you have some managed models available. You can also bring in Azure AI Foundry models and use them for your prompt.

On the panel to the right of Instructions, select the dropdown menu in the Model field, and select the plus sign (+) to connect to a model from Azure AI Foundry.
In the Connect a model from Azure AI Foundry screen, enter information about the model from Azure AI Foundry / Model catalog. Make sure to add the Model deployment name and Base model name exactly as they appear in Azure AI Foundry.
Select Connect.

Use your model

You added your model to the Model dropdown menu. Select this model for your prompt and add it to the agent. Each time this prompt runs, it uses the selected model.

Screenshot of the 'Model' dropdown menu.

These models also work well with image and documents input. However, some models have a limitation. When you add image input, the dropdown menu shows only models from Azure AI Foundry that work with images. This allows you to select a model where you can be successful when you use images or documents.

Screenshot of the 'Models' dropdown menu with image and document models.

The following models work with images:

Phi-3.5-vision-instruct
Phi-4-multimodal-instruct
Phi-3-vision-128k-instruct
o1
GPT-4o
GPT-4o-mini
GPT-4
GPT-4.5-preview

Note

Copilot Studio doesn't natively support image generation or expose Azure AI Foundry models directly in the user interface. The Add an AI model catalog currently includes AI Builder models and selected Azure AI Services, but no text-to-image models.

You can integrate image generation by using Azure AI Foundry models (for example, DALL·E 3) through a plugin or custom action. Deploy the model in Azure AI Studio, expose it as a REST API, and call the endpoint from Copilot Studio to return and display the generated image (such as a link or image card).

Supported models and limitations

Keep the following requirements and limitations in mind when you connect an Azure AI Foundry model to a prompt.

Use a chat completion endpoint

This feature supports models with the chat completion type. When you connect a model, the Azure model endpoint URL must point to the chat completions endpoint, which ends in /chat/completions. For example:

https://<your-resource>.services.ai.azure.com/openai/deployments/<deployment-name>/chat/completions

If you use the model's Responses API endpoint, which ends in /openai/v1/responses, the connection fails with a Resource not found error. Change the URL to the chat completions endpoint to resolve the error.

GPT-5 and later models aren't supported

The GPT-5 family and later models (such as GPT-5, GPT-5 mini, GPT-5 nano, and the GPT-5.x versions) aren't currently supported for bring your own model in prompts.

Manage your Azure AI Foundry model

The Azure AI Foundry models are connected by connectors. You can set the governance policy for this connector in the policy page in the Power Platform admin center under the name 'Azure AI Foundry'. This allows you to add a specific data policy for these models and the tenant data policy.

Each model where a maker set up a connection is also available as a connection page.

Screenshot of the 'Connections' page in Power Apps.

This currently supports models with chat completion type.