AI foundry responses API returns 404 for conversations that worked some times ago

Hi @Anton Zhukov,

Thanks for reporting this. A 404 NotFound error such as “Conversation ID conv_ not found in project foundry-@AML”** typically means the request reached the Foundry service, but the specified conversation ID could not be resolved within the targeted project context.

Here are a few checks to help narrow this down:

1. Verify the Foundry project endpoint Ensure you are using the Azure AI Foundry project–scoped endpoint, for example:

https://<account>.services.ai.azure.com/api/projects/<project-name>/openai/v1/conversations/...

Using an Azure OpenAI (*.openai.azure.com) endpoint or omitting the /api/projects/... path can result in a 404.

2. Confirm the conversation exists in the project Try listing conversations via GET …/openai/v1/conversations and verify that the conv_… ID appears exactly as referenced (IDs are case-sensitive).
3. Check API version and region alignment

• Make sure the required api-version (for example, v1) is specified.
• Confirm the hostname and region (e.g., swedencentral, francecentral) match the Foundry project where the conversation was created.

4. Test with a newly created conversation If the issue only affects older conversation IDs, create a new conversation and verify whether API calls succeed for it. This helps rule out ID-specific or state-related issues.

If the problem persists after these checks, please consider opening or following up on a support request and include:

• Full request URL (without credentials)
• Conversation ID
• Project name, region, and API version
• Timestamp and request_id from the error response

This information will help the backend team investigate why the conversation lookup is failing.

Hope this helps clarify the issue.

The 404 not_found error indicates that the specific conversation resource ID no longer exists or is not accessible in the target Foundry project, even if it worked previously.

From the Conversations REST reference, the GET {endpoint}/openai/v1/conversations/{conversation_id} and related operations return 404 when the requested conversation cannot be found. The error payload you see is consistent with this behavior:

• The service is successfully reached (so endpoint and auth are valid).
• The API is specifically reporting that the conversation with that conversation_id is not found in the given project.

Given that this affects conversations that used to work, the most likely explanations supported by the documentation are:

1. The conversation resource has been deleted or expired, so subsequent calls to retrieve it now return 404.
2. The request is now targeting a different project or endpoint than when the conversation was originally created (for example, a different Foundry project name or environment), so the ID is unknown in the current project.
3. The conversation_id being passed is incorrect (truncated, copied from another environment, or changed), which will also result in 404.

To move forward using the documented behavior:

1. Verify the endpoint and project
   • Confirm that the endpoint used in the Conversations API calls matches the Foundry project where the conversation was originally created.
   • Ensure the api-version is correct (defaults to v1 if not specified).
2. Confirm the conversation ID
   • Check that the conversation_id in the URL is exactly the one returned when the conversation was created.
   • Ensure there are no formatting changes (extra characters, URL encoding issues, or truncation).
3. Handle 404 as a terminal state
   • The reference for similar resources (Datasets, Indexes) states that a 404 is returned when the specific version does not exist. The same pattern applies to conversations: if 404 is returned, the resource should be treated as missing and not recoverable via the API.
   • Application logic should treat this as a signal to create a new conversation instead of retrying the same ID.

If the same conversation_id is confirmed correct and the endpoint/project are unchanged, but 404 persists across multiple regions and deployments, this indicates the conversation resources are no longer present. In that case, the only supported path is to create new conversations and adjust client code to tolerate 404s for old IDs.

References:

• Azure OpenAI in Microsoft Foundry Models v1 REST API reference
• How to generate text responses with Microsoft Foundry Models
• Microsoft Foundry
• Foundry Models frequently asked questions
• Error Handling
• Are Anthropic models deployed via Azure AI Foundry officially supported with third-party tools such as HAI or Cline? - Microsoft Q&A
• Owin security could not handle callback "signin-oidc", Error 404 On Asp.net Mvc! - Microsoft Q&A
• Azure Deployment Takes 70 Minutes to Fail with Obscure Error - Microsoft Q&A
• Problem Viewing thread. - Microsoft Q&A
• Deployed Claude and getting API deployment resource is not ready even though it has been an hour - Microsoft Q&A