Assistant: "Regenerate" button returns the previous response

tdemin

When I press "Regenerate" on an LLM response, I seemingly get responded with cached generated text instead of a new generation.

Steps to reproduce:

Open Kagi Assistant and a fresh chat.
Try "Say something random" as the prompt. The model generates an answer.
Press "Regenerate".
The response is getting returned instantaneously, the same response you've already seen.

OS: macOS 26.3/iOS 26.3, reproduces on Safari
Models: observed with at least Claude 4.6 Sonnet and GPT-5.2

RoxyRoxyRoxy

Try changing the prompt slightly and see if it occurs (even just something like removing a period). My first thought is this is likely caching but I'm not an AI engineer 🙂