I want to report a critical and fully reproducible bug that is affecting my paying customers and has generated 10+ complaints in the last 72 hours.
WHAT IS HAPPENING
The model generates a complete response, but it is NEVER displayed to the user in the public-facing deployment. The user sees the loading indicator (floating dots) indefinitely and has no reason to believe the response has been generated. Only when the page is manually reloaded does the full response appear — already stored, already complete.
This means: the model is working fine. The bug is in how the public frontend handles the streamed response.
EXACT REPRODUCTION STEPS
Open any published Pickaxe app via its public URL
Submit any message
Observe: loading dots appear and spin indefinitely
Wait as long as you want — no response ever appears
Manually reload the page
Observe: the FULL response is already there, generated and stored
CONFIRMED BEHAVIOR IN BUILDER
The exact same app, same model, same prompt tested in the BUILDER PLAYGROUND works perfectly — instant response, no streaming issues, no hanging.
This confirms the bug is ONLY in the public deployment layer, not in the model connection or server logic.
TESTING DONE
Tested across ALL available models: Gemini 3 Flash, GPT-4.1, GPT-4.1 mini, Grok 4 Fast — same behavior on all of them
Tested across 3+ different apps — same behavior on all
Fully reproducible every time
The Gemini API tested directly via Google AI Studio with the same prompts responds in seconds — the issue is not at the API level
WHEN IT STARTED
This began coinciding exactly with your recent UI/platform redesign update. Before that update, everything worked correctly.
Hi @gastroseo, thanks for posting! We also received your email about this - we have had one of our engineers working on it, and will get a fix for you today. I will follow up here (and via email) shortly!
Thank you so much for the incredibly fast response — I truly appreciate it. It’s a huge relief to know your engineering team is already on it.
This issue has been directly impacting my paying customers over the last few days, so having a fix today would mean the world to me and to them.
I’ll be standing by. Please don’t hesitate to reach out if you need any additional details, screen recordings, or access to specific apps to help reproduce the issue.
Thank you again — this is exactly the kind of support that makes the difference.
@gastroseo We figured it out! The issues were happening because you had toggled on OpenClaw sometime last week. We added it briefly last week before locking it and enabling it on an individual basis. We turned yours off manually, and things are running fine now. So sorry for the confusion, and please let us know if you have any other issues!
Glad the responses are flowing again! OpenClaw is our internal automation stack for running assistants inside the Pickaxe dashboard. We briefly exposed the toggle while we were testing, but it isn’t generally available yet, which is why flipping it on caused the embed viewer to misbehave. We’re polishing the docs and gating before we roll it out in beta—once it’s ready for creators we’ll announce it in Releases & Features and loop you in directly.