Responses api calls are not being stored like Completions for OpenAI models in Foundry

Jakhongir Mustafaev 60 Reputation points
2025-12-22T18:43:45.5+00:00

Hello,

OpenAI suggests using the Responses API over Completions; however, Responses API calls are not being stored as Completions can in Foundry.

Can the feature to also store the Responses API be added to Foundry?

Thanks!

Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
{count} votes

Answer accepted by question author
  1. Sridhar M 4,320 Reputation points Microsoft External Staff Moderator
    2025-12-22T21:20:20.6033333+00:00

    Hi Jakhongir Mustafaev

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

    OpenAI is steering developers toward the newer Responses API because it unifies chat, tool‑calling, multimodal input, and reasoning into a single structured interface. However, Azure AI Foundry’s Stored Completions feature is still architected around the Chat Completions API, not the Responses API. This is confirmed by documentation showing that Stored Completions are created only through /chat/completions calls and are visible in the Foundry “Stored Completions” pane. [How to use...soft Learn | Learn.Microsoft.com], [azurefeeds.com]

    Although the new Responses API is stateful, it does not integrate with Foundry’s stored‑completions logging pipeline. The Foundry documentation and the Responses API guidance describe its features stateful execution, tool support, message IDs but do not include any mechanism that stores those interactions in Foundry’s Completions datastore. This is consistent with current product behavior: only Chat Completions produce Stored Completions in Foundry. [Azure Open...soft Learn | Learn.Microsoft.com], [github.com]

    Developers have asked why Responses API calls are not appearing in Stored Completions, and the official answer is that this capability does not exist today. Both Microsoft Q&A and GitHub discussions highlight that even with store: true, Responses API interactions do not surface under Stored Completions in the portal, and the community is treating this as a feature request. [learn.microsoft.com], [github.com]

    Foundry does support for storage today:

    For now, the only guaranteed path to have Azure AI Foundry automatically capture, view, search, and export conversation history is to use the Chat Completions API with store: true. The underlying REST surface is /chat/completions with APIs to list, retrieve, and manage stored completions, which is separate from the Responses API endpoint (/responses). Therefore, if your workflow depends on Foundry’s “Stored Completions” UI or export pipeline for dataset creation, evaluation, or distillation, Chat Completions remains the supported route. [platform.openai.com]

    Workarounds available today

    If you want to adopt the Responses API but also preserve conversation logs, you must implement application‑side storage (database, blob + metadata, Cosmos DB, etc.). Alternatively, you can enable Azure API Management + Azure Monitor logging to capture prompts and responses for downstream analysis or evaluation. This is a documented approach for producing exportable datasets even when Foundry’s native Stored Completions pipeline does not support a given API. [learn.microsoft.com]

    Is the feature coming to Foundry

    There is no official documentation or roadmap stating that Responses API calls will be added to Stored Completions. The only signals so far come from community requests asking Microsoft to add this capability. Since Foundry engineering monitors these public channels, the best path is to upvote the Q&A thread or GitHub issue to increase visibility, but there is no confirmed timeline. [github.com]

    Reference:

    Azure OpenAI stored completions & distillation (preview)

    Create an Azure OpenAI resource

    Quickstart: Get started with Azure OpenAI Service

    I Hope this helps. Do let me know if you have any further queries.

    Thank you!


0 additional answers

Sort by: Most helpful

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.