When something goes wrong

A document is stuck in queued

Nothing is processing it. Either no ingestion worker is running, or the message broker is not configured.

This is a deployment state, not a lost document: the job exists and will run when a worker appears. See Deploying.

A document says unsupported

Primer could read the file but found nothing to index. The detail says which:

• ocr_required — no text layer, and image reading is switched off.
• no_text_found — images were read as well, and there was still nothing.
• unsupported_media_type — the format is not one Primer parses.

An upload was rejected

Code	Meaning
unsupported_content	The extension and the contents disagree, or the format is not supported
quota_exceeded	Larger than the configured upload limit
validation_failed	Missing or unusable filename
not_found	The library does not exist, or is not yours

A document says failed

A stage failed after exhausting its retries. The detail carries a short, safe explanation; the full reason, including any stack trace, is in the worker logs, correlated by job id.

Common causes are an embedding endpoint that is unreachable and a document that takes longer to convert than the configured budget.

Everything returns 404

If every library and document reports "not found", the request is arriving as a different user than you expect. Primer derives your identity from headers set by the authenticating proxy; if the proxy is misconfigured, or the app is reached directly rather than through it, requests arrive as somebody else — or as nobody.

Reaching Primer directly bypasses authentication

Primer trusts its identity headers. It must not be reachable except through the proxy, or anyone can set those headers themselves. See Identity and access.