Fix Microsoft Foundry Classic Portal Issues: Full Guide
Why This Is Happening
If you've opened the Microsoft Foundry classic portal and found yourself staring at a blank left pane, missing hub-based projects, a toggle that doesn't seem to do anything, or a complete inability to locate your Azure OpenAI resources , I've been there. I've seen this exact confusion on dozens of enterprise setups, especially when teams migrate from Azure AI Studio or Azure AI Foundry and suddenly discover that the platform has been reorganized in ways nobody bothered to explain clearly in the migration email.
Here's the core problem: Microsoft Foundry now ships as two separate portals, the "classic" version and the new version, and they are not interchangeable. They look similar enough to fool you, but they serve different purposes, expose different resource types, and have different navigation models. The new Foundry portal only shows Foundry projects. The classic portal is where everything else lives: hub-based projects, Azure OpenAI resources, and features like prompt flow or managed compute model deployments that haven't landed in the new experience yet.
What makes this genuinely frustrating is that Microsoft's in-portal error states don't tell you why a resource is missing, they just show nothing, leaving you to wonder whether you've lost your data, whether permissions broke, or whether the portal is down. It's none of those. Nine times out of ten, you're looking at the wrong portal version entirely.
The second most common cause I see: the platform has been through a significant rebrand. If you worked with this toolchain before 2025, you knew it as Azure AI Studio or Azure AI Foundry. The hub + project model, the Azure OpenAI resource pane, the Assistants API configuration surface, all of that has moved. The classic portal preserves the legacy hub-based project workflow, but getting there requires explicitly toggling away from the new experience.
Other issues that bring people here include SDK endpoint confusion (the old multi-package approach with azure-ai-inference versus the new unified project client), RBAC problems blocking access to the Management Center, and left-pane customization getting out of sync across different projects. There's also the question of preview feature visibility, in enterprise environments, admins sometimes accidentally expose or hide capabilities by misconfiguring custom roles or tags.
I know this is frustrating, especially when you're mid-project and suddenly can't find your deployments or thread storage. The good news: none of this is data loss. Everything is still there. You just need to navigate to the right surface. Let's fix it. Browse all Microsoft fix guides →
The Quick Fix, Try This First
Before going through the full diagnostic steps, try this one thing first, it resolves the majority of "missing resources" and "wrong view" issues in under 60 seconds.
Look at the top banner of whatever Foundry portal you're currently in. There is a toggle labeled "New Foundry". It's easy to miss because it sits in the top navigation bar alongside the account and subscription selectors. If that toggle is switched ON, you are in the new Foundry portal, and you will only see Foundry projects. Hub-based projects, Azure OpenAI resources, and classic features like prompt flow won't be visible here.
Turn the toggle OFF. The page will reload and you'll land in the Microsoft Foundry classic portal. Your breadcrumb should now read something like "Microsoft Foundry (classic)" rather than just "Microsoft Foundry." Your hub-based projects, Azure OpenAI resources, and all the legacy tooling should now be visible in the left pane.
One important behavior to know: whichever portal version you last used becomes the default next time you visit. So if a colleague shared a link to the Foundry portal and they were using the new version, the link will open the new version for you too. You'll need to manually toggle back each session until Microsoft persists this preference more reliably.
After switching to classic, go to the left pane and check the breadcrumb navigation at the top. Click it to see a dropdown of your recent resources, this is how you quickly jump between your hub-based projects without having to search from the home screen. If your project still doesn't appear, it may be pinned under the "...More" menu at the bottom of the left pane. That menu surfaces everything that's been unpinned from the primary navigation.
If toggling to classic still doesn't show your resources, your Azure subscription or RBAC permissions may be the issue, move on to the step-by-step section below.
Open the Microsoft Foundry portal in your browser. Look at the very top of the page, the banner area. You'll see a toggle control labeled "New Foundry". This is the single most important control in the entire portal for users who work with hub-based projects or Azure OpenAI resources.
If the toggle is enabled (blue/on), you're in the new Foundry experience. This view is intentionally limited, it shows only Foundry projects, and it's designed for teams building multi-agent applications with the new unified project model. Everything else is hidden by design, not by accident.
Click the toggle to turn it off. Wait for the portal to reload. You should now see the classic portal, identifiable by the full breadcrumb navigation at the top showing your subscription, resource group, and hub or project hierarchy.
Once in the classic portal, use the breadcrumb at the top to navigate between resources. The breadcrumb doesn't just show your current location, it's interactive. Click any level of the hierarchy to see recent resources of that type, or click the "all resources" link to get a full list. This is the fastest way to locate a hub-based project that isn't showing in the left pane.
You should confirm you're in the right place when the left pane shows sections like Define and explore, Build and customize, and Observe and improve, these are the three development stages the classic portal organizes itself around, and they don't appear in the new portal view.
Hub-based projects are the legacy project type from the Azure AI Studio era. They're distinct from Foundry projects, and only the Microsoft Foundry classic portal shows them. If you've been working in the new portal and wondering where your older projects went, this is why.
After switching to the classic portal, look at the left pane. If you don't see your hub-based project listed, click the "...More" option at the very bottom of the left pane. This reveals all the items that have been unpinned from the primary navigation. Find your project there and click the pin icon to add it back to the main left pane.
It's also worth knowing that the left pane configuration is per-project and per-user. If a colleague shows you their screen and their left pane looks different from yours while in the same project, that's intentional. Each user maintains their own left pane layout for each project separately. So if someone else pinned a feature you need, you have to pin it yourself.
To pin or unpin items: select "...More" at the bottom of the left pane, then hover over any item to reveal the pin/unpin toggle. Pin the sections you use regularly, I'd recommend keeping Deployments, Prompt flow, and Evaluations pinned if you're doing model development work in the classic portal.
If your hub-based project still doesn't appear anywhere, verify your subscription context. Click the subscription selector at the top of the portal and confirm you're scoped to the right subscription. Projects live within subscriptions, and if you've recently been added to a new subscription or had RBAC changes applied, you may need to sign out and sign back in to refresh your token.
One of the most common Microsoft Foundry classic portal setup problems I hear about is Azure OpenAI resources simply not showing up. You know the resource exists, you created it, you have the endpoint URL, you're paying for it, but the portal shows nothing.
The classic portal is the only surface that exposes Azure OpenAI resources alongside Foundry and hub-based projects. So first: confirm you're in the classic portal (toggle off as described in Step 1). Then navigate to the Management Center within the classic portal. This is the governance and administration hub, and it has a dedicated resource list that shows all resource types within your scope, including Azure OpenAI resources that might not be surfaced in the main project navigation.
If the Management Center itself is not visible in your left pane, you may not have the required RBAC role. Azure OpenAI resource access in the classic portal typically requires at minimum the Cognitive Services Contributor role or equivalent. Ask your subscription admin to verify your role assignment:
az role assignment list \
--assignee your-email@domain.com \
--scope /subscriptions/YOUR_SUBSCRIPTION_ID \
--output tableIf you have access but the resource still doesn't appear, try switching to the classic portal and then navigating directly to the resource via the breadcrumb dropdown. Click the top-level subscription name in the breadcrumb, then select "All resources" and filter by type Microsoft.CognitiveServices/accounts. This bypasses the left pane filtering and should show your Azure OpenAI resource directly.
Once the resource is visible, you can work with it in the classic portal's full feature set, including model deployments, fine-tuning configurations, and the Assistants API surface (now referred to as Agents v0.5/v1 in the classic context, noting that the new Responses API / Agents v2 lives in the new portal).
Even within the classic portal, you can create and work with Foundry projects, not just hub-based ones. The key distinction: Foundry projects give each developer self-serve access to create isolated environments for prototyping, while hub-based projects are better suited to shared team workflows with centrally managed compute and data connections.
To create a new Foundry project in the classic portal, use the Build and customize section in the left pane and look for the New project button. Give the project a name, associate it with the correct Azure resource group, and select your region. The project will provision with its own file storage, thread storage (conversation history), and search indexes as isolated units, so developers on your team can experiment without stepping on each other's data.
Once the project is created, the Overview page of that project is where you'll find your endpoints and API keys. This is important: Microsoft rotated how endpoint URLs are structured in the migration from Azure AI Services to Foundry. If you're using older SDK code that references multiple separate endpoints (one for inference, one for embeddings, etc.), you'll want to migrate to the unified project endpoint approach. The Overview page shows this unified endpoint.
For teams that need to bring their own Azure resources into a Foundry project for compliance reasons, for instance, a bring-your-own storage account to keep sensitive data within your tenant boundary, that configuration is also handled during project setup. Look for the "Advanced" options panel during project creation and configure your custom Azure resource connections there.
After setup, navigate back to the Management Center to verify the project appears in your resource inventory and that team member access roles have been applied correctly. This is also where quota management lives, if your deployments are hitting token-per-minute limits, this is the pane to check first.
If your code was working against the old Azure AI Services or Azure AI Studio endpoints and is now throwing authentication errors, 404s, or "resource not found" responses after the Foundry migration, the SDK and endpoint configuration is almost certainly the culprit.
The old approach used multiple separate Python packages and multiple endpoints:
# OLD, multiple packages, multiple endpoints (pre-Foundry migration)
from azure.ai.inference import ChatCompletionsClient
from azure.ai.ml import MLClient
from openai import AzureOpenAI
client = AzureOpenAI(
azure_endpoint="https://YOUR_RESOURCE.openai.azure.com/",
api_version="2024-02-01"
)
The new Foundry SDK model uses a single unified project client and a single project endpoint. If you're targeting the new Responses API (Agents v2), your code should use the unified azure-ai-projects 2.x client. However, and this matters, if you're working in the classic portal with hub-based projects or Azure OpenAI resources, the AzureOpenAI client still works. Just make sure your API version string is current. Old monthly api-version parameters like 2023-12-01-preview have been deprecated in favor of the v1 stable route pattern (/openai/v1/).
For the new unified SDK approach against a Foundry project endpoint:
# NEW, unified project client (azure-ai-projects 2.x)
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
client = AIProjectClient(
endpoint="https://YOUR_PROJECT.services.ai.azure.com/",
credential=DefaultAzureCredential()
)
After updating your endpoint and credential configuration, test the connection before re-running your full application. A quick connectivity check against the /openai/v1/models route will confirm whether the authentication and routing are working correctly. If you still get a 401, verify that your managed identity or service principal has the Azure AI Developer role on the Foundry project resource.
Advanced Troubleshooting
If the five steps above haven't resolved your Microsoft Foundry classic portal configuration issues, the problem is likely in one of three places: RBAC/role assignments, preview feature visibility controls, or browser-level state. Here's how to work through each.
RBAC and Management Center Access
The Management Center in the classic portal, where quota, user management, and resource governance live, requires specific roles to even see in the left pane. If it's not appearing for you, your role may not include management plane permissions. Have your subscription admin run:
az role assignment list \
--assignee YOUR_OBJECT_ID \
--all \
--output tableYou'll need either Owner, Contributor, or a custom role that explicitly grants Microsoft.MachineLearningServices/workspaces/read on the workspace scope. Viewer-level roles don't expose the Management Center.
Disabling or Showing Preview Features via Custom RBAC
Enterprise teams sometimes want to block preview features in production Foundry environments. The official mechanism is a combination of resource tags and custom RBAC deny assignments. To hide a preview-only feature, you apply a deny assignment on the specific resource action associated with that feature at the resource group or subscription scope. Microsoft's documentation on disabling preview features in Foundry describes this pattern, but be careful: deny assignments are additive and can't easily be undone via the portal. Always apply them in a test subscription first.
If you're on the other side, a developer who can't access a feature that should be available, and your admin has applied preview restrictions, check the Activity Log for your resource group:
az monitor activity-log list \
--resource-group YOUR_RG \
--start-time 2026-04-14T00:00:00Z \
--query "[?authorization.action contains 'deny']" \
--output tableThis will surface any deny assignment events that might be blocking your access.
Browser State and Token Cache Issues
The Foundry portal is a single-page application and it caches authentication tokens aggressively. If you've recently changed tenants, switched subscriptions, or had your account permissions modified, the portal can get into a state where it's using a stale token against the new permission set, resulting in blank resource lists or "you don't have access" banners even when you actually do.
Fix this by doing a hard session clear: open browser developer tools (F12), go to Application > Storage, and click "Clear site data" for the Foundry portal domain. Then sign out from the Microsoft account picker and sign back in. This forces a fresh token acquisition against your current role assignments.
Prompt Flow and Managed Compute Not Showing
These two features are explicitly only available in the classic portal, they haven't been ported to the new Foundry experience yet. If you can't find Prompt Flow in the left pane of the classic portal, it's almost certainly been unpinned. Click "...More" at the bottom of the left pane and look for it there. If it's not in the "More" list either, verify that your project type is hub-based, Foundry projects in the classic portal don't expose prompt flow the same way hub-based projects do.
Prevention & Best Practices
Once you've fixed the immediate Microsoft Foundry classic portal issue, the real goal is making sure you don't spend another hour debugging the same problem six weeks from now. Here's what I'd put in place for any team using the classic portal in a production or near-production context.
Document which portal version each workflow requires. Keep a simple internal wiki page or team README that specifies: hub-based project workflows → classic portal; new multi-agent application builds → new Foundry portal. The moment a new developer joins the team and opens the wrong portal version, they'll waste real time. This two-line note prevents that entirely.
Assign roles proactively, not reactively. The single biggest source of "where did my resources go" issues in enterprise Azure environments is a user whose role was provisioned correctly at project setup but whose access drifted because they were added to a new subscription without the corresponding Foundry role assignments. Implement a policy that ties Foundry project membership to your IdP group memberships so RBAC tracks identity changes automatically.
Pin your left pane correctly on day one. Since the Foundry classic portal left pane is per-user and per-project, new team members get a default pane that may not expose the features they need. Add a one-time onboarding step where new developers pin their relevant sections (Deployments, Evaluations, Prompt Flow, etc.) before they start building. This prevents the "I can't find the feature" issue entirely.
Test SDK connectivity as part of your CI pipeline. If your application code references Foundry or Azure OpenAI endpoints, add a lightweight health check to your CI pipeline that validates endpoint reachability and authentication token acquisition. This catches endpoint URL drift or credential rotation issues before they hit production.
- Bookmark the classic portal URL directly so you always land in the right version without toggling
- Use
az account set --subscription YOUR_SUB_IDin your local dev environment to avoid subscription context confusion when running CLI commands against Foundry resources - Set up Azure Monitor alerts on your Foundry resource group so you're notified immediately if a resource becomes unhealthy rather than discovering it when a deployment call fails
- Review quota limits in the Management Center monthly, hitting a token-per-minute cap looks like an application error, not a quota problem, and it's an easy thing to miss until it bites you in production
Frequently Asked Questions
What is Microsoft Foundry classic portal and how is it different from the new Foundry portal?
The Microsoft Foundry classic portal is the version of the Foundry platform that supports hub-based projects, Foundry projects, and Azure OpenAI resources all in one place. The new Foundry portal only shows Foundry projects and is designed for teams building multi-agent applications using the latest APIs. The practical difference: if you need to work with hub-based project workflows, use prompt flow, configure managed compute model deployments, or access Azure OpenAI resources in the portal UI, you must use the classic version. Toggle between them using the "New Foundry" toggle in the portal banner, whichever version you last used becomes your default.
How do I trace my application and monitor generative AI apps using the Foundry classic portal?
Application tracing and monitoring in the Microsoft Foundry classic portal live in the Observe and improve section of the left pane. Pin it if it's not visible by clicking "...More" at the bottom of the left pane. From there you can access tracing tools to debug your agentic workflows step by step, run evaluations to compare how different model configurations or prompts affect output quality, and connect to monitoring dashboards that track your generative AI application's performance over time. For production monitoring, Microsoft also integrates with Azure Monitor, you can route Foundry telemetry to a Log Analytics workspace for custom alerting and longer-term retention.
My hub-based projects are missing after I opened the Foundry portal, where did they go?
They're almost certainly still there, you're most likely looking at the new Foundry portal, which only shows Foundry projects. Find the "New Foundry" toggle in the top banner of the portal and turn it off. After the page reloads you should be in the classic portal and your hub-based projects will reappear. If they still don't show up after switching, check your subscription selector to make sure you're scoped to the right Azure subscription, then use the breadcrumb navigation at the top to browse all resources and filter for your project type.
How do I get started with AI training and certification for Microsoft Foundry?
Microsoft offers a dedicated AI learning and community hub with multiple paths relevant to the Foundry platform. The most relevant tracks are "Develop generative AI apps in Microsoft Foundry," the AI Engineer career path, and the module on responsible AI principles. These are available through Microsoft Learn and can be accessed at no cost. For teams, the AI learning hub also has organizational learning paths that an admin can assign across a team. If you're working specifically with Foundry agents and agentic workflows, the "Develop generative AI apps in Microsoft Foundry" module covers project setup, model deployment, and the new Responses API in practical hands-on exercises.
What's the difference between Threads/Assistants API and the new Conversations/Responses API in Foundry?
This one trips up a lot of teams mid-migration. The Threads, Messages, Runs, and Assistants terminology is from the old Assistants API (Agents v0.5/v1), it's what you worked with if you built agents on the classic portal surface before the Foundry rebrand. The new terminology, Conversations, Items, Responses, and Agent Versions, belongs to the Responses API (Agents v2) and is the current recommended approach for new builds. The classic portal still supports the old API surface for backwards compatibility, but new agent projects should use the Responses API via the unified project endpoint. Your existing code using Assistants API patterns will keep working for now, but plan to migrate before the deprecation window closes.
How do I disable preview features in the Foundry classic portal for a production environment?
Microsoft provides two mechanisms for this in the Foundry classic portal. The first is resource tagging: you apply specific Azure resource tags to your Foundry resources that the portal reads to hide preview-gated features from that resource's project view. The second is custom RBAC deny assignments: you create a custom role that denies the specific resource action associated with each preview feature, then assign that deny role to the production resource group. The deny assignment approach is more surgical but harder to reverse, so test it in a non-production subscription first. You can also use the Management Center's organization controls to limit which feature tiers are available across a set of resources, this is the preferred approach for large enterprise environments with many projects.