Hey,
I've been a user of Kagi Assistant and really enjoy it.
Nevertheless I've found three bugs that impact my usage.
Let's say I use Claude for the initial few messages of a conversation and then edit one of the user messages to resend them with another LLM selected. When i hit send, it reverts my selection to the LLM used before and runs the inference with it. This mostly happens with larger conversations, for shorter ones I can switch just fine. For me it happened especially often when using R1 and trying to switch to Claude.
Especially in older conversations that I've saved, the <> buttons to switch between the different branches of a conversation glitch. Sometimes I get shown no reply at all when I hit them, sometimes I can just go back 1 branch and then nothing happens, and sometimes nothing happens at all when i click the < button. For me this also happens in other browsers and when sharing the conversation.
This is a smaller one and a more minor inconvenience. When using thinking models such as R1, I want to read the thinking as it happens and also already during the inference process. This means I have to expand the thinking section manually as soon as it appears, which is already not ideal. But then what happens during the main inference is that I have to click on the thinking section again and again after each new section in the inference or wait until the LLM is finished, so I can read the thoughts, as the thoughts section is automatically collapsed after each section in the main inference text.
Extra suggestion:
Maybe allow to name the branches in a way that <1/4> can become < MyText> for example
- Always run the inference with the LLM currently selected in the drop down menu.
- Being able to use < and > to navigate responses
- Being able to have the thinking section exposed during the inference process. When I click on it, I want it to stay exposed, even though the inference may still be ongoing.
