Our Thunk appears to be malfunctioning. Rerunning all records/workflow steps did not resolve the problem. It basically gets stuck “Waiting” at Step 2 (“Collect information from URLs”) where the “Execution AI” area either (a) it gives us a “progress update” where it says it’s processed 5 URLs and still has to process the remaining URLs, or (b) displays a message “There was a problem with the LLM” followed by a bunch of HTML.
Hi @Rachel there was a Cloudflare outage that affected Thunk this morning. It should be resolved https://www.cloudflarestatus.com/
Can you try restarting the step(s) that were affected? On our side things seem okay now.
Hi @tony – rerunning – I’ll get back to you in a few!! Thanks!!!
Hi @tony – the problem persists 
Can you link me to a step where this is happening + is it okay if I run it myself to reproduce the issue? Thanks!
Hi Tony - do you have access to our Thunk? If you do, would it be easier for you to just go in? All records trying to run are affected, and it’s always Step 2. I can’t remember how to send URLs to steps… and yes, it’s OK for you to run whatever you want.
Some error messages I’m seeing:
Using multi_tool_use.parallel is not allowed with functions that are not available. Retrying with allowed tools.- DO NOT RUN. Placeholder to satisfy parallel tool structure if needed.
- Note: Tool set does not expose update_item directly under functions
- There was a problem with the LLM: 500
I think most of these are just the Thunk talking, but the “ There was a problem with the LLM: 500” is consistent.
I have access to your Thunks, but I see multiple associated with your account. The ones I look at have no rows in the table view, so I’m not sure which one you’re looking at. Could you paste the link for me, then I can take a look? Thanks.
Oh yes - sorry - I created a bunch the other day because I was testing something (I want to delete all of those but can’t figure out how). The Thunk is “EBSCO Title Search v3” Thunk
1 Like
Thank you, @Rachel got it.
Let’s divide and conquer…
The “ There was a problem with the LLM: 500” error is related to the Cloudflare outage. I spoke too soon about it being over - we’ve been getting hit with more of those.
Now onto these:
Something weird is happening here. Your steps are executing using GPT-5, even though you and Jason both a different model set. We haven’t thoroughly integrated GPT-5 into our system and we see some strange behaviors with it.
Also - this might account for your timeouts, since GPT-5 is significantly slower than the other models.
We are looking into this and will get back to you.
Hi @tony – I had set the model to GPT-5… I will revert it to Default… I wonder why it’s not showing you that I set it to that? What is it showing I set it as?
When I looked at the step instructions, they showed Default. But both your and Jason’s default models were set to gpt-4.1-mini, which is a bit puzzling and makes me think there’s a bug.