Microsoft Foundry: Agent Creation Failing with HTTP 404 in East US 2

Question

Microsoft Foundry: Agent Creation Failing with HTTP 404 in East US 2

MrJohnson-5267 0

Problem description

I am unable to create agents in my Azure AI Foundry project located in the East US 2 region. All attempts to create new agents, including minimal prompt-only agents with no tools, result in an HTTP 404 not_found error, stating that a referenced resource was not found in the target project or workspace.

Environment

Azure AI Foundry project under an AI Services account in East US 2. Affected capability: agent creation in the project. Surfaces: Foundry portal and REST API for agent creation and version creation.

What I've already tried

I identified the issue after agent creation started failing on 2026-10-28. I retried agent creation through multiple paths: using the Foundry portal, POST /agents, and POST /agents/{name}/versions via REST with a new agent name. All paths showed the same HTTP 404 not_found error. I also reduced the scenario to a minimal prompt-only agent with no tools; the failure persisted. Additionally, I assigned the Foundry User role to the project and observed no change in behavior. I verified role assignments and account configurations, including system-assigned managed identity, custom subdomain, and allowProjectManagement setting.

Current status

As of 2026-10-29, agent creation in the project continues to fail with the same HTTP 404 error in both the portal and API, even for minimal agents. I am seeking assistance to examine the backend state of the project and resolve this issue, as the failure has persisted despite role assignments and configuration checks.

Karnam Venkata Rajeswari 5,340 Reputation points Microsoft External Staff Moderator

2026-10-29T20:15:46.9666667+00:00
Hello @MrJohnson-5267 ,

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

Since the same HTTP 404 appears in the portal and the REST API, even for a minimal agent with no tools, the behavior suggests that the project is not able to resolve a resource it needs at the time of agent creation, rather than an issue with the prompt, tools or permissions.

This usually comes down to one of two things: a project-level state issue on the service side or a missing dependency, such as the model deployment referenced by the agent.

Please check if the following steps help -

Run a read test on the affected project. Copy the endpoint from the project home page (format: https://{account-name}.services.ai.azure.com/api/projects/{project-name}) and list the agents: az rest --method GET --url "https://{account-name}.services.ai.azure.com/api/projects/{project-name}/agents?api-version=v1" --resource "https://ai.azure.com" Please note whether this succeeds and whether an existing agent can be retrieved. This shows if only creation fails or the whole project is affected.

Please confirm that the model deployment named in the agent definition still exists on the Foundry resource, and retry with another deployment if one is available.

Create a second project under the same Foundry resource in East US 2 and try a minimal prompt agent there. Please keep the original project and do not delete it, as it will be needed for the investigation. If the new project works, the issue is limited to the original project, and the new one can serve as a temporary workaround with the agent definitions recreated there. If it also fails, the issue is more likely at the resource or regional level. Creating the project from the Foundry portal is preferred, as the required roles are added automatically. If CLI or SDK is used, Foundry User needs to be assigned manually.

Without changing anything, verify that Foundry User is assigned on the Foundry resource to both the user principal and the project's managed identity.

If own storage resources (Cosmos DB, Storage, AI Search) were configured for agents, please confirm they still exist, are reachable and were not restricted

The following references might be helpful , please check them out –

· Quickstart: Create a prompt agent - Microsoft Foundry | Microsoft Learn

· Create a project - Microsoft Foundry | Microsoft Learn

· Role-based access control for Microsoft Foundry - Microsoft Foundry | Microsoft Learn

· Manage hosted agents - Microsoft Foundry | Microsoft Learn

· Azure Service Health Portal - Azure Service Health | Microsoft Learn

We have reached out to you on private message for further assistance

Thank you
MrJohnson-5267 0 Reputation points

2026-11-02T20:35:09.63+00:00
Thanks for the steps. Results from 2026-11-02:

Read test: succeeds. GET /agents?api-version=v1 returns the project's three existing agents, and each can be retrieved by name. The existing agents also still run. Only creation fails.

Deployments: gpt-5.4 and gpt-5.2 both exist with provisioning state Succeeded, both in ARM and from the project's /deployments endpoint. Creation fails with either one.

Second project: we created a new project in the same Foundry resource from the portal and tried a minimal prompt agent on gpt-5.4. It [succeeded / failed with the same 404, request ID ...]. The original project is unchanged.

Foundry User: the user creating agents has Foundry User directly on the Foundry resource and through its resource group. The project's managed identity doesn't have it. Neither does the managed identity of our prod project (same subscription and region), where creating agents works, so we left it unchanged as you suggested.

We don't use our own storage for agents: there are no capability hosts on the resource or the project. The only connected resource is an Azure AI Search service, which is running and reachable, and the existing agents use it.

Creation still fails as of 2026-11-02 20:26 UTC, request ID 461e3366b70528c37859820c0ad69cab.

One more detail for the investigation: the agent identity blueprints Foundry created in this project carry a virtualWorkspaceId tag that doesn't match the project's internal ID. In our working prod project the two match. I can share the exact values and resource IDs privately.

Your answer

MrJohnson-5267 0 Reputation points

2026-11-02T20:35:09.63+00:00

Thanks for the steps. Results from 2026-11-02:

Read test: succeeds. GET /agents?api-version=v1 returns the project's three existing agents, and each can be retrieved by name. The existing agents also still run. Only creation fails.

Deployments: gpt-5.4 and gpt-5.2 both exist with provisioning state Succeeded, both in ARM and from the project's /deployments endpoint. Creation fails with either one.

Second project: we created a new project in the same Foundry resource from the portal and tried a minimal prompt agent on gpt-5.4. It [succeeded / failed with the same 404, request ID ...]. The original project is unchanged.

Foundry User: the user creating agents has Foundry User directly on the Foundry resource and through its resource group. The project's managed identity doesn't have it. Neither does the managed identity of our prod project (same subscription and region), where creating agents works, so we left it unchanged as you suggested.

We don't use our own storage for agents: there are no capability hosts on the resource or the project. The only connected resource is an Azure AI Search service, which is running and reachable, and the existing agents use it.

Creation still fails as of 2026-11-02 20:26 UTC, request ID 461e3366b70528c37859820c0ad69cab.

One more detail for the investigation: the agent identity blueprints Foundry created in this project carry a virtualWorkspaceId tag that doesn't match the project's internal ID. In our working prod project the two match. I can share the exact values and resource IDs privately.