Validate the install¶
A few checks confirm that configuration, the metadata store, the embedding provider, the queue, and the retrieval path all work.
1. Print the resolved configuration¶
Validate settings and print them. Pydantic masks secret values in this output, so it is safe to share when reporting a problem:
uv run --env-file .env python - <<'PY'
from refindery.config import load_settings
print(load_settings().model_dump_json(indent=2))
PY
If this raises, a setting is invalid — the error names the offending field. See the Settings reference.
2. Liveness and readiness¶
After starting the server, check the two health endpoints. They are
unauthenticated. Liveness (/healthz) says the process is up; readiness
(/readyz) additionally confirms the metadata store is reachable and an
embedding model is active (503 until then).
3. End-to-end check¶
Run the ingest and search requests. A page that reaches
indexed and a search that returns it exercises the whole path: queue →
embedding provider → vector store → retrieval.
If readiness never turns green
The most common cause is entity extraction having no healthy extractor at
startup. Install the ner extra or configure another extractor chain — see
Entities and Operations.