Degraded Performance Issue. KB Not Fully Executing, Severe Latency

Posting this in case others are experiencing the same and to get some visibility from the Pickaxe team.

Started seeing degraded performance on our end around 11:00 AM PST yesterday. Here is what we observed:

  1. Slow, laggy response (typing out each word very slowly, ~5 seconds per sentence).
  2. Knowledge base was not being fully executed. The model was partially following instructions but not completing the full required sequence. No changes were made to our KB or model configuration on my end, so this appeared to be platform-side.
  3. AI summary in user activity was showing N/A

The output and KB issue seemed to resolve around 1:00 AM PST. However, performance is still not good.

Current symptoms as of this morning:

  1. GPT-5.4 mini (fast setting) is taking up to 50 seconds to think per turn
  2. Output is extremely slow, typing out each word very slowly, ~5 seconds per sentence

Would appreciate any visibility from the Pickaxe team on whether this is a known ongoing issue and what the current status is.

2 Likes

hi @next-era !

thanks for asking the right questions !

this is indeed a known issue with a few OpenAI models over the last day or so.

re: KB & conversation summaries, both of these can be attributed to OpenAI latency streams to the API as well.

yesterday, we computed the conversation summaries by hand when we noticed that the job wasn’t triggering. as you stated, both of these seem to be less of an issue currently.

with that being said, we have been iterating on some changes over the last 24 hours and changes are still actively coming to improve/harden known latency hotspots.

how are things looking today on your end ? i have just done a few preliminary tests on my end on both my own agents/agents in your workspace and things are looking a little better.

talk soon,

carson