Azure Foundry using fabric ontology is returning server error

Hello @Avadh Kishore ,

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

Since the returned message is a generic server-side error, the underlying exception is not exposed in the portal and can only be confirmed through backend diagnostics associated with the provided Request ID.

One important consideration is that this scenario uses a Fabric Ontology (Fabric IQ) Knowledge Source, which follows a live retrieval model and is different from traditional ingestion-based or indexing-based knowledge source scenarios. Because retrieval occurs against live Fabric resources, authentication, runtime access, ontology configuration, and tenant alignment become the highest-priority validation areas.

Please check if the following validation steps help-

1. Validating Runtime Access and Authorization Since Fabric Ontology retrieval occurs at query time, please confirm that the runtime identity has access to:
   • The Fabric workspace.
     • The Fabric Ontology.
       • The Direct Lake semantic model.
         • The underlying Fabric data required to answer the query.
   Although workspace administration and Managed Identity permissions have already been configured, successful resource creation does not always guarantee successful runtime retrieval access. This is one of the most important validation steps for Fabric Ontology scenarios.
2. Verifying Tenant and Resource Alignment Please verify that:
   • Azure AI Foundry, Azure AI Search, and the Fabric workspace belong to the same Microsoft Entra tenant.
     • The Azure AI Search connection remains healthy and verified.
       • No firewall, private endpoint, or networking restrictions are preventing runtime communication.
   Tenant alignment is a documented prerequisite for Fabric Ontology Knowledge Sources
3. Reviewing the Fabric Ontology Configuration Please validate that:
   • The ontology is healthy and available.
     • Entity definitions, properties, relationships, and measure mappings are correctly configured.
       • The ontology metadata is aligned with the Direct Lake semantic model.
         • Data bindings for the measures being queried are complete.
   Incomplete ontology metadata or bindings can result in retrieval failures or unexpected runtime behavior
4. Confirming Knowledge Base Retrieval Settings For Fabric Ontology Knowledge Sources, Minimal Retrieval Reasoning Effort is not supported. The recommended setting is Low or Medium. Please review the current Knowledge Base configuration and update the setting if necessary before retesting.
   1. Performing Isolation Testing
   To determine where the failure originates, the following tests are recommended:
   1. Query the Fabric Ontology directly from Fabric.
      1. Create a new test agent and attach the same Knowledge Base.
         1. Test simple ontology prompts such as:
            • “What entities are available in this ontology?”

              - “List available measures.”
              
                 1. Create a simple non-ontology Knowledge Base (for example, using a PDF document) and test it with the same agent.
              
                    1. If semantic model reasoning is involved, retest using a supported latest-generation model.
              
                    These tests help determine whether the issue is isolated to the ontology, Knowledge Base, agent configuration, or runtime retrieval layer.
              

The following references might be helpful , please check them out

• Configure Logging - Azure AI Search | Microsoft Learn
• Fabric data agent creation - Microsoft Fabric | Microsoft Learn
• Connect agents to Microsoft Fabric with Fabric IQ (preview) - Microsoft Foundry | Microsoft Learn
• Use the Microsoft Fabric data agent with Foundry agents - Microsoft Foundry | Microsoft Learn
• Agent identity concepts in Microsoft Foundry - Microsoft Foundry | Microsoft Learn

Please let us know if the response was helpful

Thank you

Hello @Avadh Kishore ,

Following up to see if you had any chance to review the above response

Thank you