Hi I encountered an issue with the AI agents executing tool call twice when the initial tool call is taking too long and I try to say or generate reply inside the tool call: "AI state changed: listening -> thinking" AI executes tool 1 Tool 1 has some delay After delay threshold reach, I trigger say or generate reply to say the filler. The AI say the filler AI executes tool 1 again I think this happens because the tool call is not added to the chat context until the result arrives? would this be correct? I can see some users reporting this issue as well.

There was a similar thread, where the issue was resolved by adding

tool_choice="none"

to the

generate_reply

call that is happening within the function tool

Tried this and didn't work for me tho. This doesn't stop the AI to output a tool call if I generate reply. I think the best solution is if I can add the tool call to chat context even without the result yet. Cause seems like its only added to chat context after tool returns.

We have seen something similar, out of curiosity what model are you using @able-airline-55422?