Update: the root cause was an incident at our hosting provider, Railway ("Domain routing disruption", https://status.railway.com/incident/IFBXJTGG). Railway's report shows 07:48 UTC because it was posted after the fact. The outage itself ran from about 07:35 to 07:40 UTC, as described above.

Between approximately 07:35 and 07:40 UTC on September 30, new connections to app.manifest.build received HTTP 404 errors and never reached Manifest. This affected the dashboard and API traffic, including LLM requests routed through the gateway. For about one more minute after traffic returned, some requests were slow or timed out.

Root cause: a network configuration change rolled out by our hosting provider made its global routing layer answer "not found" for hosted applications in every region. The provider's system recovered by itself. Manifest's application and database stayed up throughout and did not restart, and no data was lost.

Requests that failed during this window can be safely retried. Service has been fully operational since 07:40 UTC.