Reasoning traces are leaking into ai assistant responses to clients

Community Discussion Bug Report
1

i have recently changed the LLM behind the AI assitant to Gemini 3.5 Flash Lite. Since that changes we have now seen 2 cases where the (english) reasoning traces are leaking into the chat response.

  • is this caused by tawk.to?
  • can you prevent this from happening?
  • would changing to a different LLM help?
2

Hi Thijsjan,

Thanks for bringing this to our attention.

Could you please email me these chat transcripts (kristaps@tawk.to) so we can troubleshoot?