The updated assistant often fails to call the ResearchAgen

JBum

In many of my conversations with Kagi Assistant with the recent update, the models end their text output with:

Let me call the ResearchAgent to verify current best practices for this setup.

And they seem to fail to actually call the research agent.

Happy to provide any context or example conversations where necessary.

The assistant should actually query the ResearchAgent to get a search result.

Rehan

One or two examples would be great. Just checking, this is with the regular assistant, and not a custom assistant?

JBum

Hi @Rehan, thanks for getting back.

Assistant Info

This is not with the regular assistant, it's a bare-bones custom assistant using Sonnet 3.5. The assistant currently has no special instructions. I created it almost purely to have a shorter search bang operator. See the assistant config:

Example Link

Here is a link to a conversation where the assistant did this in the first message.

https://kagi.com/assistant/50f3da03-f726-4cc7-8ab3-2c0b48759ee8

Specifically, the final paragraph is where this happens:

Let me call the ResearchAgent to verify if there are any recent changes or best practices for systemd socket activation with Docker containers.

You can see from the chat linked that this was a long message, so I'm imaging some token limit was hit, preventing the actual request.

I hope this is useful.