Errors - Valkyrie

Valkyrie uses standard HTTP status codes. The table below covers the cases you’re most likely to hit and what to do about them.

Status	Meaning	What to do
`200`	Success	Nothing to do
`401`	Missing or invalid credentials	Check your `X-API-Key` / bearer token
`403`	Authenticated but not allowed	You don’t have access to this resource; check model sharing
`404`	Not found	Check the deployment slug or alias in the URL
`409`	Conflict	e.g. registering an email that already exists
`429`	Rate limited	Back off and retry with jitter
`5xx`	Server or upstream error	Retry with backoff; if it persists, check the deployment’s health

Handling a stopped deployment

If a deployment is stopped, inference calls will fail until it’s resumed. Resume it from the Endpoints screen (or with the MCP tools), then retry.

Debugging

Verify the base URL and model/slug are correct.
Confirm the API key is valid for this environment (keys are environment-specific).
Check the deployment’s health if inference calls fail intermittently.
For fine-tuning issues, stream the job logs to see where a run stopped.

Build retries with exponential backoff and jitter for 429 and 5xx. Treat 4xx responses other than 429 as caller errors, retrying won’t help until you fix the request.

Inference API

​Handling a stopped deployment

​Debugging

Handling a stopped deployment

Debugging