Hi Team, I'm trying to use claygent (running GPT 5 mini) to search the web. It has previously worked before, but is now failing on almost all tries, with the messages: "Search tool failed repeatedly; unable to locate or verify any public statements. Returning blank per instructions." "Unable to perform Google searches due to tool failures; cannot locate a direct quote. Returning empty response per instructions." Any idea what's going wrong here? Table - https://app.clay.com/workspaces/223473/workbooks/wb_0t3cyp95mGNnCro9WNH/tables/t_0t4u52mEcreN98Djp8H/views/gv_fN4U5XMkE74k
This looks like an API tier issue. Claygent web research requires OpenAI Tier 2 or higher (≥450,000 TPM) to function properly. [1]
Check your OpenAI API tier at https://platform.openai.com/settings/organization/limits. If you're on Tier 1 or free tier, you'll need to upgrade to Tier 2.
Also verify your API key is correct at https://platform.openai.com/account/api-keys and update it in Clay if needed. Rate limits can cause these search tool failures, especially with recent updates that are more conservative with token usage.
Click into the specific failing cells to see more detailed error messages that might help pinpoint the exact issue.
Was that helpful?
Nope - I'm Tier 5
I understand this is frustrating, so I’m routing you to one of our human agents who can assist further.
Clay Support is sharing a file with you.
Our support team has got your message and we'll get back to you soon! We're currently outside of our standard office hours (9 AM to 9 PM EST, Monday through Friday), so it may take a little longer for us to respond.
If you’re dealing with a specific table, drop the URL below so we can help you quicker. Otherwise, someone from our team will be in touch soon!
Hi Joey, thanks for reaching out! Sorry to hear about this. When you mentioned that it previously worked before, did it work with the same prompt? I'm wondering if the workflow would be more accurate if we used 'Perform Search' using Google --> and then used Claygent to pull out the quotes from the result that the google search finds.
Hi Natalie, yep it worked before in other tables - and I use GPT or Sonnet all the time to do web based research, curious why it would be failing now.
Thanks Joey! Do you mind sending over the tables where it worked before? Would love to look at the logs and compare how they're different in these two. Thank you!
Natalie H. - sure, https://app.clay.com/workspaces/223473/tables/t_0t1539p7ZzEm8qjMWQY/views/gv_fN4U5XMkE74k It's Column "2-B" (2nd column on the left). My new table (which is an exact copy) = https://app.clay.com/workspaces/223473/workbooks/wb_0t3cyp95mGNnCro9WNH/tables/t_0t4u52mEcreN98Djp8H/views/gv_fN4U5XMkE74k Looking through the old table, there doesn't seem to be any of these types of failures. However, when I run some of the rows in the old table again - it gives the new error/failure I'm receiving in the new table - so somethings changed.
Also, when setting my API key - it's giving the below error (I'm Tier 5 - so not sure why this issue would show) "Could not finish the message because max_tokens or model output limit was reached. Please try again with higher max_tokens."
Thanks Joey! I see what you're saying about how the old table has a different message. I took a look at the "Steps Taken" that Claygent was doing in your old "Ecommerce V2 - Research Table" and it wasn't looking at the LinkedIn profile, and instead was using google search. So this may be why it was able to go and find quotes:
For your second question about the max_token error in your API key, sorry for this, it looks like we've got a bug on our hands. I've escalated this to our engineering team to help look into it. We'll follow up here as soon as we know more!
