Custom Model Adding through Admin Center Is possible into Copilot studio

An administrator cannot inject an arbitrary internal model into the agent’s primary Model dropdown, but a maker can connect a supported Azure AI Foundry deployment to the separate prompt-level Model dropdown.

Two different model dropdowns

1. Agent primary model
   • Located at Agent → Overview → Model.
   • Controls the model used for the agent’s orchestration.
   • Contains Microsoft-supported internal and external models available for the environment, region, and enabled administrator policies.
   • Administrators can allow or block preview, experimental, and supported external models, but the documented controls do not register arbitrary endpoints in this picker.^1^
2. Prompt-level model
   • Located inside a Prompt tool.
   • Supports bringing an Azure AI Foundry model into that prompt’s Model dropdown.
   • The connected model runs only when that prompt executes; it does not become the agent’s primary orchestration model.^2^

Add an Azure AI Foundry model to a prompt

For a model deployed through Azure AI Foundry:

1. Open the agent in Copilot Studio.
2. Select Tools → Add a tool → New tool → Prompt.
3. Enter the prompt name and instructions.
4. In the right pane, open Model and select +.
5. In Connect a model from Azure AI Foundry, enter:
   • Azure model endpoint URL
   • Model deployment name
   • Base model name
   • Required connection credentials
6. Enter the deployment and base-model names exactly as shown in Azure AI Foundry.
7. Select Connect, select the newly connected model, and add the prompt to the agent.

The endpoint must be a chat completions endpoint ending in /chat/completions. A Responses API endpoint ending in /openai/v1/responses fails with Resource not found. GPT-5-family and later models are currently not supported by this prompt-level BYOM feature.^2^

Conceptually, the supported endpoint format is:

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

What the admin centers control

The Power Platform and Microsoft 365 admin centers govern access; they do not provide a general-purpose Add custom model to primary picker function.

Administrators can:

• Allow or block preview and experimental models.
• Enable cross-region data movement where required.
• Enable supported external providers such as Anthropic, Mistral, or xAI.
• Apply a data loss prevention policy to the Azure AI Foundry connector used by prompt-level models.^1^^2^

Environment feature settings are under:

Power Platform admin center → Manage → Environments → select environment → Settings → Product → Features

Updating these settings requires the System Administrator security role or equivalent permissions.^3^

If the model is an internal or custom endpoint

If it cannot be deployed as a supported Azure AI Foundry chat-completion model:

• Expose it through a secured API.
• Invoke it from the agent through a custom connector, custom action, plugin, or another supported tool integration.
• Return a structured response to the agent.

In that design, the model remains an agent tool or action. It does not appear as the agent’s primary model or replace Copilot Studio’s orchestration model.

Therefore, the conclusion is:

• Primary agent Model dropdown: no arbitrary internal/custom-model injection.
• Prompt Model dropdown: supported Azure AI Foundry deployments can be added.
• Other internal endpoints: integrate as tools/actions, not native primary models.
• Admin centers: enable, restrict, and govern supported models and connectors; they do not onboard arbitrary models into the picker.

References

1. Select a primary AI model for your agent
2. Bring your own model for your prompts
3. Manage feature settings