[QUESTION] Why is the OpenAIClient class's GetResponsesClient() function experimental? Any ETA to remove the experimental flag?

[kunalk16]

How can we help?

Was wondering why the OpenAIClient class's GetResponsesClient() function is still experimental and if there is any ETA when we are going to have it removed? If the experimental flag is removed, we can update downstream repos like azure-sdk-for-net also to remove the experimental flag from AzureOpenAIClient. As of now, my team is building something using Azure's nuget and folks are hesitating to use the responses client due to this flag.

[jsquire]

Hi @kunalk16. Thanks for reaching out and for your inquiry. The short version is that [Experimental] is used to indicate that the client API surface for a feature area is still undergoing changes and does not offer guarantees of backwards compatibility from release-to-release. Currently, there is no timeline to share about when Responses will progress to a non-experimental state, but I can say that it will apply to the entire feature area when that happens.

More details on our use of experimental can be found below.

What's Experimental used for?

The short version is that Experimental is used to designate that the client API surface is still undergoing iteration and should be expected to change, potentially in ways that are breaking. This may include both compile-time and behavioral changes in the client API. To acknowledge this risk, .NET requires that you opt-into experimental features by suppressing the warning in whatever way you prefer.

Once a client API reaches a stable state (no longer tagged experimental) then it is covered by a strong backwards compatibility guarantee. We will not make binary breaking changes and will only make disruptive behavioral changes when there is a critical need, such as patching a security vulnerability.

What Experimental isn't

• Client use of Experimental does not communicate any message about the feature in the REST API. The OpenAI platform docs are the authoritative source for that information.

• Client use of Experimental does not communicate any message about the quality or reliability of the feature. We believe that all client features that we release meet our quality bar and are safe and reliable to use in production environments.

• Client use of Experimental does not communicate any message about supportability of the feature. We support all released client features when used with the official OpenAI service.

Additional context

• Feature Lifecycle in the OpenAI .NET Client Library

[kunalk16]

Thanks for the information @jsquire !