Models retrieve are not the deployed ones.

[lumimario]

Which component is this issue related to?

Other / Not sure

Which Umbraco AI version are you using? (Please write the exact version, example: 10.1.0)

v1.2.0

Bug summary

This is related to Umbraco.AI.MicrosoftFoundry v1.1.1

The models that appear in the Model dropdown when creating a profile seems to be all the available in Foundry and not just the deployed ones.

Specifics

Umbraco:

Deployed in Foundry:

Steps to reproduce

1. Create a connection to MS Foundry
2. Create a Context and select the created connection
3. Select Model. Here you can see that it displays every single model, even when they haven't been deployed.

Expected result / actual result

I just want to see my deployed models in the Foundry project.

Dependencies

No response

[mattbrailsford]

Investigation Summary

Root Cause

The provider currently calls GET {endpoint}/openai/models?api-version=2024-10-21 which returns all models accessible by the Azure AI resource (the entire catalog/base models + fine-tuned models), not just the ones deployed to the user's project.

Microsoft does not provide a data plane endpoint for listing deployments using API key authentication. The old GET {endpoint}/openai/deployments?api-version=2023-03-15-preview endpoint was deprecated and removed.

Available API for Deployed Models

The Azure AI Projects data plane API provides a deployments listing endpoint:

GET {endpoint}/deployments?api-version=v1

This is available via the Azure.AI.Projects .NET SDK (projectClient.Deployments.GetDeployments()), but it requires Microsoft Entra ID authentication (DefaultAzureCredential) — it does not support API key auth.

The API key is still required for actual LLM inference calls (chat completions, embeddings etc.), so Entra ID would only be used for the model discovery/listing operation.

Proposed Fix

Add an optional Entra ID authentication configuration to the Microsoft AI Foundry provider settings:

• If Entra ID auth is configured: use the /deployments?api-version=v1 endpoint to list only deployed models
• If not configured: fall back to the current /openai/models behavior (all available models)

This keeps the provider backwards-compatible — users who only have an endpoint + API key get the current experience, while users who enable Entra ID get a filtered list of their deployed models.

References

• Deployments - List REST API (AI Foundry)
• Azure.AI.Projects .NET SDK
• Q&A: Azure AI Foundry API endpoint for listing model deployments

[mattbrailsford]

Correction: The REST API reference link in my previous comment is a dead link on Microsoft's side. Here are the working references:

• Azure.AI.Projects .NET SDK - Deployments operations
• Azure.AI.Projects NuGet package
• Q&A: Azure AI Foundry API endpoint for listing model deployments

The key .NET SDK code for listing deployed models:

AIProjectClient projectClient = new AIProjectClient(new Uri(endpoint), new DefaultAzureCredential());

foreach (AIProjectDeployment deployment in projectClient.Deployments.GetDeployments())
{
    Console.WriteLine(deployment);
}

This calls GET {endpoint}/deployments?api-version=v1 under the hood (data plane, requires Entra ID auth).

[mattbrailsford]

I've added a known limitation to the README for now https://github.com/umbraco/Umbraco.AI/tree/dev/Umbraco.AI.MicrosoftFoundry#known-limitations

[lumimario]

Thanks Matt, sounds good.

[mattbrailsford]

If you are happy to run from source I do have a PR here #84 which updates the MS Foundry provider to use their newer SDK and also uses Entra ID for accessing models which then allows only fetching deployed models. It currently falls back to the previous behaviour though if Entra ID auth fails, or if you are using the older API Key based access.

[lumimario]

Thanks Matt, yes, I might give it a try.

[mattbrailsford]

An updated MS Foundry provider has been released with Entra ID support which now only lists deployed models.