Document Intelligence Studio message "Failed to access Document Intelligence"

Question

I have a new account in Document Intelligence. I was able to access projects, do some tutorials, and then started doing some closer to real work tutorials using the same format as the tutorial. In all cases, I was working with the "Custom Extraction Model".

The next day I came in and all my models are not working. The actual project opens with all the label data I entered correctly displayed, but I get an error message with "Failed to access Document Intelligence", Failed to send request to Document Intelligence. Please check your Document Intelligence service settings and network connection.

I can add additional fields on the screen. My real issue starts whenever I do an action that leaves the Label Data screen. For instance, if I click the "Train" button, I can fill in the details, but when click the actual "Train" or "Test" action buttons, I get the same error and the training fails.

Answer 1

Hello @Steven Banks [E] (VAC/ACC) ,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

Based on the observed behavior, the issue appears to occur when Document Intelligence Studio performs service-side operations such as Train and Test, where communication with the Document Intelligence resource is required. Since the project remains accessible and labeling information is still available, the current evidence indicates that the project data is preserved and recreating the project or relabeling documents is not expected to resolve the issue.

Please check if the following steps in the mentioned order help-

1. Capturing the failed request details from browser Developer Tools The first step is to determine whether the request is being blocked before reaching Azure or whether the request reaches the service and receives an error response. Steps:
   1. Open Document Intelligence Studio.
   2. Press F12 to open Developer Tools.
   3. Select the Network tab.
   4. Enable Preserve log.
   5. Reproduce the issue by selecting Train or Test.
   6. Locate the failed request and capture:
      • Request URL.
      • HTTP status code.
      • Error message.
      • UTC timestamp.
      • Request ID/correlation ID, if available.
      The result will determine the troubleshooting path:
   If there is no HTTP status code Examples:
   • Failed to fetch.
   • net::ERR_FAILED.
   • Blocked request.
   • CORS or browser-related errors.
   This usually indicates an issue with:
   • Network connectivity.
   • Firewall/proxy restrictions.
   • VPN routing.
   • Browser security policies.
   • Private endpoint or public access configuration.
   If HTTP 401 or 403 is returned This indicates that the request reached Azure but access was denied. Review:
   • Authentication configuration.
   • Azure RBAC permissions.
   • Resource access restrictions.
   If HTTP 404 is returned Review:
   • Selected Document Intelligence resource.
   • Resource endpoint.
   • Subscription and resource configuration.
   If HTTP 500 or 503 is returned Review:
   • Azure Resource Health.
   • Azure Service Health.
   • Request identifiers for further investigation.
2. Verifying the Document Intelligence resource configuration Please confirm that Document Intelligence Studio is connected to the expected Azure resource. Review: Azure Portal > Document Intelligence Resource > Overview Validate:
   • Correct tenant.
   • Correct subscription.
   • Correct resource group.
   • Correct Document Intelligence resource.
   • Correct region.
   • Resource status is healthy.
   Also confirm that the endpoint information has not changed.
3. Reviewing Document Intelligence networking configuration Since the issue started after previously working successfully, networking configuration should be reviewed as a priority. Azure Portal: Document Intelligence Resource > Networking Verify:
   • Public network access is enabled.
   • Firewall rules have not changed unexpectedly.
   • Selected network restrictions are not blocking access.
   • Private Endpoint configuration and DNS resolution are correct, if private networking is enabled.
   For troubleshooting, temporarily test with public network access enabled and retry the Train/Test operation.
4. Validating browser, proxy, VPN, and enterprise network access Security policies or network changes can affect service communication without changes being made to the Document Intelligence project. Recommended checks:
   • Test without VPN connectivity.
   • Test from an alternate network, such as a mobile hotspot.
   • Test using another supported browser.
   • Use an InPrivate/Incognito browser session.
   • Temporarily disable browser extensions.
   If the operation succeeds from another network, review corporate network rules and proxy configuration for access to:
   • *.cognitiveservices.azure.com
   • documentintelligence.ai.azure.comSecurity policies or network changes can affect service communication without changes being made to the Document Intelligence project. Recommended checks:
   • Test without VPN connectivity.
   • Test from an alternate network, such as a mobile hotspot.
   • Test using another supported browser.
   • Use an InPrivate/Incognito browser session.
   • Temporarily disable browser extensions.
   If the operation succeeds from another network, review corporate network rules and proxy configuration for access to:
   • *.cognitiveservices.azure.com
   • documentintelligence.ai.azure.com
5. Reviewing authentication and permissions (if indicated by HTTP 401/403) f the Network trace shows authentication-related errors, validate access permissions. Azure Portal: Document Intelligence Resource > Access Control (IAM) Confirm appropriate role assignments:
   • Cognitive Services User.
   • Cognitive Services Contributor (where required for management operations).
   If managed identity-based access is configured, validate the associated permissions.
6. Validating storage access for custom model training Storage validation is recommended if the request reaches the service successfully but training fails during processing. Verify:
   • Storage account still exists.
   • Container is available.
   • Training documents have not been deleted or moved.
   • SAS tokens are valid, if used.
   • Storage permissions are configured correctly.
   • Storage networking restrictions are not preventing access.
7. Performing an isolation test To determine whether the issue affects the entire resource or only the existing custom extraction projects:
   1. Create a new Custom Extraction project.
   2. Upload sample documents.
   3. Add labels.
   4. Attempt training.
   Results: If the new project also fails: The issue is likely related to:
   • Resource configuration.
   • Networking.
   • Authentication.
   • Service availability.
   If the new project succeeds: Further review can focus on:
   • Existing project configuration.
   • Dataset-specific dependencies.
   • Storage references.
   Recreating existing projects or relabeling documents is not recommended because the current evidence indicates that the existing labeling information remains available.To determine whether the issue affects the entire resource or only the existing custom extraction projects:
   1. Create a new Custom Extraction project.
   2. Upload sample documents.
   3. Add labels.
   4. Attempt training.
   Results: If the new project also fails: The issue is likely related to:
   • Resource configuration.
   • Networking.
   • Authentication.
   • Service availability.
   If the new project succeeds: Further review can focus on:
   • Existing project configuration.
   • Dataset-specific dependencies.
   • Storage references.
   Recreating existing projects or relabeling documents is not recommended because the current evidence indicates that the existing labeling information remains available.

The following references might be helpful , please check them out

• Quickstart: Document Intelligence Studio - Foundry Tools | Microsoft Learn
• Configure Virtual Networks for Foundry Tools - Foundry Tools | Microsoft Learn
• Authentication in Foundry Tools - Foundry Tools | Microsoft Learn
• Create Document Intelligence Studio custom projects - Foundry Tools | Microsoft Learn
• Authorize Blob Access with Microsoft Entra ID - Azure Storage | Microsoft Learn

Please let us know if the response was helpful

 

Thank you

 

Please "Accept" the answer with an "Upvote" if the response was helpful. This will be benefitting other community members who face the same issue.

 

Answer 2

Hi, because the project and labels still open but Train/Test calls fail, check the live service dependency rather than rebuilding labels. Confirm the Document Intelligence resource still exists and the Studio project points to the correct endpoint/key or Entra identity. Then verify your Cognitive Services role, Storage Blob Data Contributor access on the linked container, CORS/network/firewall/private-endpoint settings, and whether local authentication was disabled. Use the browser network response and resource diagnostic logs to distinguish 401/403 from DNS/network failure. If nothing changed in configuration and all projects failed overnight, capture the request/correlation ID and UTC time and open a Document Intelligence support case before recreating the project.

Answer 3

This error usually means Document Intelligence Studio signed in successfully, but the Studio frontend could not complete a request to the underlying Document Intelligence resource.

For the scenario described, check these areas in order:

1. Verify the bound Document Intelligence resource still exists
   • If a custom project opens but the resource that project is bound to was deleted or moved, Studio can show project data but fail when it needs to call the service.
   • A documented symptom is Form Recognizer Not Found when opening a custom project.
   • Resolution:
     • Re-create the Document Intelligence resource under the same subscription, resource group, and name, or
     • Re-create the custom project with the migrated Document Intelligence resource and use the same storage account.
2. Check permissions on the Document Intelligence resource
   • If local key-based authentication is disabled and the signed-in principal does not have enough permission, Studio can fail when opening projects or using prebuilt/custom operations.
   • A documented symptom is PermissionDenied.
   • Fix the Azure role assignments for the account being used.
3. Check permissions on the linked storage account
   • If the storage account has local authentication disabled and the required blob permissions are missing, Studio can fail when opening or operating on a custom project.
   • A documented symptom is AuthorizationPermissionMismatch.
   • Fix the Azure role assignments for storage access.
4. Confirm the correct directory and resource are selected in Studio
   • If the account has access to multiple directories or resources, switch to the correct Directory and Resource in Document Intelligence Studio.
   • In Studio, select Settings in the upper-right corner, then open the Resource tab to view or switch the active resource.
5. Check browser cookie settings
   • Studio can fail to sign in or maintain access correctly if the browser blocks non-Microsoft or third-party cookies.
   • For browser settings:
     • Edge: Settings → search third-party → Manage and delete cookies and site data → turn off Block third-party cookies
     • Chrome: Settings → search Third-party → under Default behavior, select Allow third-party cookies
     • Firefox: Settings → search cookies → Enhanced Tracking Protection → Manage Exceptions → add an exception for https://documentintelligence.ai.azure.com or the Studio URL for the environment
     • Safari: Preferences → Privacy → clear Block all cookies
6. If training specifically fails with server-side errors
   • There is a current known issue where model training can fail with HTTP 500 / InternalServerError when managed identity is not configured correctly.
   • Enable Managed Identity on the Document Intelligence resource and verify that identity has the required permissions.
7. If the model remains stuck in Not started
   • This is a documented backend lifecycle issue.
   • The backend clears failed training jobs after about seven days.
   • If earlier removal is needed, submit an Azure portal support request.
8. If the problem is intermittent
   • There has been at least one reported Studio access incident that was later mitigated.
   • If the issue started suddenly after it worked the previous day, and permissions/resource selection are correct, this can also be a temporary service-side issue.

A practical path for this case is:

1. In Studio, open Settings → Resource and confirm the expected resource is selected.
2. Verify that resource was not deleted or moved.
3. Verify access roles on both the Document Intelligence resource and the linked storage account.
4. Allow third-party/non-Microsoft cookies in the browser and retry in a fresh session.
5. If the failure happens only on Train, verify managed identity is enabled and correctly permissioned.
6. If training jobs are stuck in Not started, wait for backend cleanup or open a support request.

---

References

• Get started: Document Intelligence Studio
• Project sharing using Document Intelligence Studio
• Azure Document Intelligence known issues and troubleshooting
• Document Intelligence Studio - Microsoft Q&A
• Cannot access Azure Document Studio - Microsoft Q&A