Share via

We've built a RAG chatbot using Azure openAI assistants API. Do we need to migrate this, and what's the best option?

Ollie Gooding 40 Reputation points
2025-07-24T07:16:29.9966667+00:00

The assistants API works great for our use case- querying simple documents from a chatbot interface, linked to a few different easily managed vector stores.

However, I see the API is being mothballed in 2026. Also, it's not extending model compatibility to the frontier models. Can you confirm that is correct?

I was considering migrating to Azure open AI Responses API which has access to the more recent models, but I don't want to do this if everything except the Agent service is going to be left to fall obsolete. If we migrate to responses API, do we risk needing to migrate again within a year?

Looking at the Agent Service though, I'm struggling with the documentation a bit. The basic service seems like it would do what we need with file search, but it's not clear where the vector stores would actually be located.

The standard service seems more advanced than we need, and more expensive with Azure AI search.

Any extra info much appreciated.

Azure OpenAI Service
Azure OpenAI Service
An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
0 comments
Answer accepted by question author
  1. Jerald Felix 7,910 Reputation points
    2025-07-30T04:20:16.4266667+00:00

    Hello Ollie Gooding!
    I understand you’ve built a RAG chatbot using the Azure OpenAI Assistants API and are concerned about future support, especially as this API is scheduled to be phased out in 2026. You're also exploring migration options to avoid multiple future moves and are weighing the Responses API and Agent Service.

    Here’s a helpful overview:

    API Deprecation: Yes, Microsoft has indicated that the Assistants API will be mothballed in 2026, and it won’t be updated for newer, "frontier" models. This means staying on Assistants API may not be future-proof for advanced LLM use.

    Responses API: Migrating to the Azure OpenAI Responses API gives access to the most recent models, but there’s uncertainty whether this API, except for core Agent Service features, will continue to receive the same level of support or future updates. If you migrate now, it’s possible that Microsoft may further consolidate or evolve these APIs again after a year or two.

    Agent Service: The Agent Service is designed to provide robust orchestration for tools like file search and retrieval, and from your description, the "basic" tier could work for your needs. However, documentation about where vector stores are physically or logically managed can be unclear. Typically, for the Agent Service, Azure AI Search (or another compatible vector DB) is required, which may be more complex or costly than your current setup.

    Migration Advice: If your use case is basic RAG with manageable vector stores and minimal orchestration, it may be worth experimenting with the basic Agent Service now, as it is likelier to be Microsoft's long-term foundation for AI workflows. Keep an eye on announcements about future roadmap and deprecation schedules, as Microsoft often provides at least a year’s notice before major changes.

    What you can do:

    Reach out to Microsoft Support or your Azure account rep for definitive guidance on vector storage with Agent Service and the planned longevity of Responses API.

    Start a limited test migration to Agent Service and document any feature, pricing, or integration concerns.

    Stay flexible with your app's architecture in case another migration is needed in future.

    If you need step-by-step advice for any migration path or want further clarification on Azure’s documentation, let me know—I’m happy to help!

    Best regards,

    Jerald Felix

    0 comments

0 additional answers

Sort by: Most helpful
Most helpful Newest Oldest

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.