QueryTool Issues
# support
m
I'm experiencing an issue today where the agent goes silent after receiving a response from the "Query Tool". The agent doesn’t continue the conversation unless I say something first. Call ID: 8f63ad8a-2bf8-42ff-8661-7172cc50fdcc
c
To fix the issue, reset the tool’s messages field to an empty array (\[\]) via API to restore normal conversation flow..
r
Wouldn't that wipe out the tool's default message settings, so next time it wouldn't emit, say, "Hold on a minute" etc.? Or maybe I don't get what you're suggesting. Thanks for letting me jump in here.
c
It shouldn't say anything if the array is empty. Alternatively, you could create a request-complete message for the query tool which may restore normal conversation flow as well. Please try out the solutions provided and let us know what works and what doesnt.
l
I have the same issue, this solution really fixed it. please put this in your docs, I wasted 5 hours searching for a workaround, didn't know what was wrong, and the fact i just supplied a message for the agent to say: hmmm, let me see (before he uses the tool) doesn't really make sense.
Could you please update when it will be possible to add a message before tool lookup? It would be more natural for the agent to say: hmmm. let me see, or "just me moment I'll take a look" and then provide the answers. Right now it's just a long pause between user questions, and we really need to be able to let the user know the agent is "looking up something"
j
want to chime in here since i faced the same issue Was able to fix it by deleting the "Messages" within the query tool through the dashboard. Didn't need to patch anything through API. Hope it helps someone https://cdn.discordapp.com/attachments/1399951950883328021/1402373605073092639/image.png?ex=6893adb2&is=68925c32&hm=cb0ac7e405093cb686adf663df697246142308a9099efb910d32e3dc065d9477&
m
even after removing the message, Im facing a long delay when the agent use the query tool and when asked for services the agent state all services with description it can talk continually for a whole minute which is bad behavior despite instructing it so limit to 2-sentences to not overwhelm the customer
l
Hi, try using a different model that is more respectful for your instructions (play around with few) It's really up to the LLM to decide what parts are being sent to TTS (Text to Speech). As such this is part of your system prompt and the way it is structured and how clear it is does make a difference. (I had the same problem if reading too much, but adjusting the instructions did fix it)
c
Is everyone good here to mark the ticket as resolved for the time being?
2 Views