I have noticed that there is seemingly no option to prevent thinking/reasoning data from streaming into a Pega GenAI agent widget (Pega 25.1.2 version).
Scenario: I use Claude Sonnet 4.6, but I don’t want the agent widget to display thinking data to the user; only the actual response shall be shown.
Based on my understanding, the model endpoints themselves offer a clear separation into text and thinking blocks, so something like this should be possible.
Is there any way to achieve this programmatically at the moment in Pega 25.1.2 (you can, of course, instruct the agent to exclude this data from its response via a prompt, but that is unreliable and not a good way to do this, I’d say)?
Hello Franz I think default SSE protocol is used to stream the agent output in case of conversational widget. Worth sharing this as a feedback request to the platform.
Hi @Franz_H, at the moment, including in 25.1.2 and 26, there is no explicit configuration that allows you to include/exclude the thinking data from the Conversational Agent stream.
I understand from my colleagues in the Product department that they’re considering adding options for this in the future.
For the moment, I see good results in achieving this through the agent prompt, e.g. by including a strict output structure example.
Thanks a lot Tim for your response :)!
It would be very much welcomed if Pega could add this. I myself, unfortunately, experience only a very limited success rate with preventing that behavior via prompting.
But it’s really good to know that it can’t be done currently, thanks!