Foundry Knowledge Base Does Not Automatically Enable Private Execution Environment for Azure AI Search Indexers

[SACHIN KN (sachinkn777)]

Technical Feedback

The Microsoft Foundry Knowledge Base feature creates Azure AI Search indexers without automatically enabling the private execution environment. As a result, ingestion fails when the configured data source is accessible only through private networking. This issue is not limited to Azure Storage accounts; it also affects other Azure services configured for private access, such as Azure SQL Database and similar private endpoint-enabled resources. Even when the required Shared Private Links are configured and approved, the generated indexers continue to use the default execution environment, leading to 403 Forbidden - Public access is disabled and related connectivity failures.

Desired Outcome

Foundry should automatically detect when a Knowledge Source uses a privately accessible Azure resource and configure the underlying Azure AI Search indexer to use the private execution environment. Alternatively, Foundry should expose an option during Knowledge Source creation to enable private execution on the generated indexer.

This would:

• Ensure ingestion works seamlessly with private endpoint-enabled Azure resources.
• Eliminate manual post-deployment configuration.
• Improve scalability when managing multiple Knowledge Sources.
• Reduce operational overhead and the risk of ingestion failures in enterprise environments that enforce private networking.

Current Workaround

Describe the current workaround.
After Foundry creates the Azure AI Search indexer, users must manually update the indexer configuration and set:

JSON
"executionEnvironment": "private"

This can be done through the Azure AI Search REST API, SDK, or Azure Portal before running ingestion.

While this workaround resolves the issue, it must be repeated for every newly created Knowledge Source and every generated indexer. This approach does not scale for environments with multiple private data sources and introduces additional operational burden and configuration management overhead.