The go-live checklist

Ten minutes of checks before launch saves you a bad night.

7 minLesson 9 of 90 of 2 steps

You now know what you need. This last short lesson is a list to read once before going live.

Keys

  1. Each service and each environment (development and production) has its own key.
  2. Keys live in environment variables or a secrets store, never in code or the repository.
  3. Every key has a monthly or daily cap.
  4. Expiry dates of important keys are in your calendar.

Money

  1. The balance covers a few peak days and the low-balance alert is on.
  2. The account's email is verified and notifications reach someone who'll act.
  3. max_tokens is realistic for each kind of request.

Reliability

  1. Critical models have a fallback.
  2. 429 and 503 are retried with a growing delay, and retry-after is honoured.
  3. You write full model ids, not aliases.
  4. Requests that must not repeat send Idempotency-Key.
  5. The client timeout is long enough for slow models.

Seeing what happens

  1. You log each request's x-request-id.
  2. You send X-Title so usage shows per application.
  3. You know how to filter Usage & logs.

If you do only one thing on this list: put a cap on every key.

Team

  1. Whoever leaves is removed from the organization at once, and their keys are turned off.
  2. Guardrails are defined for sensitive data and allowed models.

Run the checklist0 of 2 steps

  1. Go through the list for one real service.
  2. For each "no", write down one small task.
The course ends here, but models and prices keep changing. Look at the blog and rankings now and then.

What you learned

  • Separate keys, caps, fallbacks and proper retries.
  • Keep the x-request-id.
  • Wherever you get stuck, open a ticket from Support.
The last lesson of the courseDone