All Systems Normal

For roughly two hours, AI replies in the dashboard editor chat and the site builder chat could finish on the server but never render in the panel, leaving the conversation looking stuck or blank. No edits or generations were lost — the work completed behind the blank message.

Affected

1. resolved · 13 Aug, 07:20 UTC

   Both chat surfaces have been rendering replies normally since the fix went out. This incident is resolved. If you still see a blank assistant message in an old conversation, reloading the page restores it.

2. monitoring · 13 Aug, 06:40 UTC

   A fix has been deployed. Replies now render in both the dashboard editor chat and the builder chat, and previously blank replies reappear in the conversation on reload. We are monitoring.

3. identified · 13 Aug, 05:50 UTC

   We traced the problem to the chat response rendering path shared by both chat surfaces: completed responses were not being committed to the message list, so the panel kept showing an empty pending state. Site generation and edits themselves were unaffected — only the display of the reply.

4. investigating · 13 Aug, 05:05 UTC

   We received reports that the AI assistant in the dashboard chat and the site builder stopped showing replies. Messages sent by users appeared, but the assistant's answer never rendered.

A provider-side update to the Moonshot AI model we use for parts of the build pipeline changed its response behaviour, causing new website generations to run long and fail instead of completing. Existing published sites were never affected.

Affected

1. resolved · 12 Aug, 23:40 UTC

   Generation success rates are back to normal levels. This incident is resolved. Any generation that failed during the window can be retried — if a failed build during this window left you short on tokens, email hello@inminutes.app and we will top your balance back up.

2. monitoring · 12 Aug, 22:30 UTC

   We raised the response limits for generation and gave every build a wall-clock budget, so a slow build now reports back with a clear message instead of dying silently. New generations are completing again and we are monitoring the success rate.

3. identified · 12 Aug, 18:20 UTC

   The failures started with a provider-side update to the Moonshot AI model used in our build pipeline. The updated model returns longer responses than before, which pushed builds past our request limits — generations hit the ceiling mid-build and errored out instead of finishing.

4. investigating · 12 Aug, 15:40 UTC

   We are seeing an elevated rate of failures on new website generations. Published sites, the dashboard, and billing are unaffected. We are investigating.