Cold starts and warm-up

Each service starts on demand and shuts down after 5 idle minutes. The next request waits while it starts again.

Two habits avoid surprises:

1. Use a generous timeout. Give your HTTP client a timeout of about 2 minutes.
2. Warm up ahead of time. When your app gets a moment of warning (a user opens a screen, starts a session, is about to upload a document), start the services you’re about to use:

call("/v1/warmup", {"services": ["decide", "language-id"]})

curl -X POST "$OLAVERSE_URL/v1/warmup" \
  -H "Authorization: Bearer $OLAVERSE_KEY" \
  -H "Content-Type: application/json" \
  -d '{"services": ["extract"]}'

/v1/warmup returns straight away and starts the services in the background. It accepts any of:

Service	Warms
decide	Decisions
extract	Document extraction
language-id	Language identification
diacritize	Diacritic restoration (diacnet models)
diactag	Diacritic restoration (diactag-2.0)
questions	Question generation