Create Preview web experience from new AI Foundry is incomplete

[Jandev]

When you create an agent in AI Foundry there's a Preview dropdown from which you can select View sample app code.

This option points the user to this repository, mentions it should clone it and create a .env file and fill it with the variables for the specific agent.

AZURE_EXISTING_AGENT_ID="My-agent:2"
AZURE_ENV_NAME="agents-playground-1234"
AZURE_LOCATION="swedencentral"
AZURE_SUBSCRIPTION_ID="C9FA0200-BF02-4F0A-AD47-ECF3138D657F"
AZURE_EXISTING_AIPROJECT_ENDPOINT="https://my-project-resource.services.ai.azure.com/api/projects/my-project"
AZURE_EXISTING_AIPROJECT_RESOURCE_ID="/subscriptions/C9FA0200-BF02-4F0A-AD47-ECF3138D657F/resourceGroups/ai-foundry-experiments/providers/Microsoft.CognitiveServices/accounts/my-project-resource/projects/my-project"
AZURE_EXISTING_RESOURCE_ID="/subscriptions/C9FA0200-BF02-4F0A-AD47-ECF3138D657F/resourceGroups/ai-foundry-experiments/providers/Microsoft.CognitiveServices/accounts/my-project-resource"
AZD_ALLOW_NON_EMPTY_FOLDER=true

From my experience, this is not complete.

This variable is also required, when you have multiple AI Foundries.

AI_FOUNDRY_RESOURCE_NAME="my-project-resource"

Based on the already provided variables, this variable does appear to be redundant as it's mentioned multiple times in the .env file already.

I was also prompted to add this environment variable:

AI_AGENT_ID="My-agent"

However, adding it to the .env file didn't help. I had to add it to the azd environment variables.

azd env set AI_AGENT_ID "My-agent"

Which is a bit weird.

Same applies to the AI_AGENT_ENDPOINT, which is being used when creating the deployment, but not mentioned in the .env file sample in AI Foundry, even though another variable is specified over there with useful information.

---

It would be great if the steps in AI Foundry are updated with the latest state of this repository. Or the repository changed so it matches with the AI Foundry steps.

[anya2975]

@Jandev were you able to get your agent working once you got the app deployed? After running into the same issues and eventually getting the app up, I could not get the agent to return a response. I'm running a custom MCP tool on my agent. I know it's an app issue because the agent works fine in preview.

[Jandev]

Interesting!

Yes, my agent is working properly now and users are working with this sample.
You might want to check the Log Stream in the Container App to see what is happening. With a bit of luck, stuff is logged over there that can help you troubleshoot.

[tmontes537]

Hello everyone! @Jandev , would you mind to share a couple of screenshots to see how the GUI looks like? Thanks!

[anya2975]

@tmontes537 I still can't get it to return a response but here's a few screenshots. It basically looks just like copilot/teams.

[Jandev]

That's indeed how the web application looks like.

If this occurs, it's probably a good idea to check if you can find something in the 'threads'-section of the Agent in Foundry. It might give a pointer on why it's failing.
Can be a dozen of things. Like ill-configured tools, instructions, knowledge bases, files that can't be accessed, etc.

I've seen also when you've upgraded a solution from the old Foundry to the new Foundry, the agents also broke. It's quite annoying, to say the least.

[tmontes537]

Thanks @Jandev ! I was curious about the GUI. I deployed mine via Corporate Teams, and it's working to some degree of decency, it's consuming a custom mcp server and an AZ index. However from the Teams environment it failed to follow the mcp tools calling route. I had to narrow it down to one tool that retrieves a bunch of data. Was wondering if it was worthy to try the web app instead. This is how it looks on Teams. BTW, it failed on copilot too, timing out after a bit.

[Jandev]

Sounds like an interesting problem. There can be a variety of things causing the behavior.
Let us know how/if you have figured it out. I'd love to read it.