Every failure on the real traffic path now says what happened and what to do next, with the code kept for support.
An error should help you take the next step
The gateway update made failed requests easier to investigate by pairing a readable explanation with an error code. Keep the code and request or trace identifier when asking for help; they identify the failure more reliably than a screenshot of the message alone.
Check authentication and scope
Confirm that the request uses the Stimulir inference endpoint, a valid application key and the intended project. A Console sign-in token and an application key have different purposes. See Authentication and workspaces and API keys.
Check the model and request
Select a model available to the workspace, check required fields, and reduce an oversized request before retrying it. A provider, quota or capacity failure needs a different response from invalid credentials. Avoid repeatedly retrying an unchanged request that cannot succeed. Follow Troubleshoot a failed request and the inference reference.
Share useful context
Open the corresponding trace when one was recorded. Note the environment, project, time, error code and trace identifier. Never include an API key, provider credential or a private request body in a public report. Use authenticated Console support for account-specific investigation.
About this entry
This expands the original 15 September 2026 announcement. Its original deployment record was not retained. Exact wording can change; use the current error code and linked troubleshooting guidance.