Früher bekannt als Azure-KI-Dienste oder Azure Cognitive Services; eine einheitliche Sammlung vordefinierter KI-Funktionen innerhalb der Microsoft Foundry-Plattform
Hi Admin Aebischer Etienne STAWI,
Thank you for sharing the detailed findings. This helps to clearly understand what is happening.
From your observation, the agent itself is still available because the GET API is working and returning the agent configuration. This means the agent is not deleted and the configuration is stored correctly.
However, the issue starts when you try to invoke the agent runtime. Both runtime endpoints are returning HTTP 404, which means the execution endpoint is not available. In simple terms, the system can see the agent definition, but it is not able to run it.
This situation usually indicates a runtime availability or routing issue, not a data issue. So the agent exists, but the system is not able to map or expose the runtime endpoint for it.
From what you described, it also confirms that the problem is likely related to the project state. Since you mentioned that the old project seems broken and you had to create a new one, it suggests that the old project may be in an inconsistent state. In such cases, it is possible that agent metadata is still present, but the runtime layer is not properly connected.
Another important point is that you are using the preview feature with the header: Foundry-Features: AgentEndpoints=V1Preview
Since these endpoints are still in preview, there can be situations where the agent configuration API works, but the runtime endpoint is not fully provisioned or exposed in that project.
The fact that both tested routes return 404 also suggests that the runtime endpoint is not registered for that agent in the project, rather than a request formatting issue.
In your case, creating a new project and being able to create agents again is a strong indicator that the issue is isolated to the old project. Continuing with the new project is a good approach.
You can also make sure in the new project that:
The agent is created and tested in the same project context
The model deployment is properly linked to the agent
The region and resource are aligned for project, model, and agent
These are required for the runtime to work correctly.
So overall, based on your findings:
Agent configuration exists and is retrievable
Runtime endpoint is not available
Cause is likely project-level inconsistency or runtime mapping issue
Using a new project resolves the issue
This is a very useful observation, and your testing clearly shows the difference between agent configuration and runtime availability.
Hope this helps. Do let me know if you have any further queries.
Thankyou!