OpenAI Response Agent Capabilities

[iFoton]

Hello,
I tried switching my agents to the recently introduced OpenAI Response Agent, but I’ve run into a few issues:

• It seems the Response Agent ignores the ResponseFormat setting in OpenAIPromptExecutionSettings and always returns plain text.
• I’m unable to retrieve logs from agent execution — including token usage and the final prompt.
• Additionally, I don’t see any agent execution logs in the OpenAI logs (https://platform.openai.com/logs?api=responses).

Is there a new way I should configure this feature, or are these capabilities just not implemented yet?

[TaoChenOSU]

Hi @iFoton,

Thank you for your question!

Could you let us know which language you're using?

[iFoton]

.net 9

[moonbox3]

Adding @markwallace-microsoft for visibility.

[CynthiaLuijkx]

Would really appreciate the ResponsesAgent to support the ResponseFormat as well!

Used Assistant previously but the latency improvement of the Responses API is so large I don't want to switch back for these kind of features.

(Also in .NET)

[itayfisz]

Hey,
It seems like OpenAIPromptExecutionSettings is totally ignored when using OpenAIResponseAgent.
When calling agent.InvokeAsync directly, it can be bypassed, as you can pass OpenAIResponseAgentInvokeOptions:

agent.InvokeAsync(... , options: new OpenAIResponseAgentInvokeOptions()
{
TextOptions = new()
{
TextFormat = ... // This will be used as responseFormat
}
}

But! When NOT using InvokeAsync directly, for example via agents orchestration (as described here), it seems it's impossible to have any impact on the parameters sent to OpenAI (reasoning effort, response format, etc). It feels that OpenAIPromptExecutionSettings should be respected in this flow as well.

@markwallace-microsoft
Could you please share if and when the Response API will be fully supported in SemanticKernel .NET? This API is a game-changer in terms of performance and costs, and brings with it additional crucial functionality. In my honest opinion, in order for SemanticKernel to continue serving greatest and latest offerings, I believe this should be prioritized.