Is it possible to use the Responses API with semantic kernel yet?

[mkemmerz]

Is it possible to use semantic kernel already with the new Responses API: https://platform.openai.com/docs/api-reference/responses?

I had some looks into the code but couldn't find anything, also this issue exists: #11026

Sorry If that question is stupid but does someone know more about the support for the Responses API? I couldn't really find anything about it.

[moonbox3]

Hello @mkemmerz it should be landing soon, in the form of an agent.

[mkemmerz]

Thanks a lot for the reply. We are looking forward to it :)

[thomasjlittle]

Just bumping this to check in, is there any news on when this might be available for .net?

[moonbox3]

Adding @markwallace-microsoft for visibility.

[readingdancer]

I'm also interested to know when support for the Responses API will be landing as I am using SK, but I'd really like to change to use this new API for the built in web search and image generation capabilities.

[moonbox3]

We have the AzureResponsesAgent/OpenAIResponsesAgent available in Python. The .NET version should be released in a few weeks.

[readingdancer]

My whole application is .NET based, the comments have been "soon" for a few months now. Do you have a nightly / pre-release that could be tested before the full release?

[markwallace-microsoft]

@readingdancer What functionality do you need in the initial release?

[readingdancer]

Currently I am using SK with the Completions API and using streaming, I have a lot of SK functions that the Agent is able to use, I currently have a web scraping function that is running in the cloud that the agent is able to call to get web content ( this is functionality that is natively supported within the Responses API, so it would be great to get rid of my dependency on another service )

I have also ( just yesterday ) written a function to allow the agent to call the image generation API, again, I believe this works much better if called from the Responses API for things like editing, using the same seed to generate similar images etc.

The Responses API allows you to stream the results of an image that is being generated so you see it progressively appear, in the same way it happens in the ChatGPT interface, this obviously gives a much better UX that telling a user they have to wait 30-60 seconds before the image appears in full.