Azure OpenAI Conversations API endpoint returns 404 on all tested configurations

Hello @administrator ,

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

The Conversations API endpoint is documented but not yet fully deployed across all environments

The POST /openai/v1/conversations endpoint is available in the official REST API reference, however current platform behavior indicates that it is not yet exposed as part of the actively deployed v1 Generally Available surface across all resources and regions.

At the same time, the API lifecycle guidance confirms that the v1 API currently supports only a subset of capabilities, with additional features continuing to roll out progressively

This explains the observed pattern:

• POST /openai/v1/responses is suppoetedand available and returns HTTP 200
• POST /openai/v1/conversations is not yet exposed which returns HTTP 404

As asked,

1. Private preview / allowlist requirement
   1. There is no official documentation indicating that the Conversations API requires allowlist enrollment.
   2. No tenant-level or subscription-level access approval is referenced.
2. Allowlist enablement possibility Since no allowlist mechanism is defined, no formal onboarding or enablement request process exists for this endpoint.
3. Feature registration or additional enablement
   • There is no documented feature registration or manual enablement step associated with /openai/v1/conversations.
   • The behavior is instead aligned with feature rollout timing across environments.

For multi-turn conversational scenarios, the Responses API can be used effectively to achieve the same outcome.

POST {endpoint}/openai/v1/responses
{
  "model": "<deployment-name>",
  "input": "follow-up message",
  "previous_response_id": "<id from previous response>"
}

This helps to -

1. Maintain multi-turn conversation context automatically
2. Eliminate the need to manage full message history manually
3. Provides functional equivalence to a persistent conversation_id workflow
4. Fully supports across current GA environments

To ensure consistency and eliminate edge conditions, the following checks can be helpful

1. Confirm usage of the correct v1 route pattern:
   • /openai/v1/...
2. Ensuring request structure follows v1 design:
   • Deployment name passed as "model" in request body
3. Validating REST execution independently:
   • Include proper headers (API key or Bearer token)
   • Capture apim-request-id for traceability
4. Collecting request diagnostics:
   • Full request URL
   • UTC timestamp
   • Response headers and payload
5. Verifying behavior across:
   • Multiple regions
   • Multiple resource types (OpenAI vs AI Services)

Since this behavior is tied to feature rollout, availability is expected to evolve over time.

The following references might be helpful , please check them out

• Azure OpenAI in Microsoft Foundry Models v1 API - Microsoft Foundry | Microsoft Learn
• Microsoft Foundry REST Reference - Azure OpenAI - Conversations | Microsoft Learn  

Thank you

Please "Accept" the answer with an "Upvote" by clicking on "Accept Answer" and thumbs-up button if the response was helpful. This will be benefitting other community members who face the same issue.