Hi everyone,
I'm building a voice-enabled solution on Azure AI Foundry that allows users to interact with Genie using speech instead of text. The architecture involves exposing a Genie Space as a tool within Foundry, which works perfectly in my own environment.
However, when deploying the same solution in a customer tenant, I consistently get a 400 Bad Request error — even when calling the Genie tool directly, without any voice layer involved.
What I've ruled out: The configuration steps are identical. The Genie Space is created and registered as a tool in Foundry without any issues.
The key difference I noticed:
- My environment: Single subscription hosting both Azure AI Foundry and Azure Databricks
- Customer environment: Two separate subscriptions — one for Azure Databricks, one for Azure AI Foundry
This is the only meaningful difference I can identify between the two setups.
My question: Has anyone encountered this issue before? Is there a known limitation or additional configuration required when Foundry and Databricks live in different Azure subscriptions? Could this be a networking/identity issue (e.g., managed identity permissions, private endpoints, or CORS policies across subscriptions)?
Any guidance would be greatly appreciated. Thanks in advance! 🙏
