Going to production
Apps/Web apps/Going to production

Going to production

Production doesn't exist until the first deploy, and a web app brings its own email, sign-in and address. This is what to set up before that deploy, what the deploy does, and what to watch once people arrive.

Before the first deploy

Decide who sends email

Without RESEND_API_KEY and EMAIL_FROM, Inference Bureau sends production's sign-in emails and team invitations for the app, in its own wording, up to 100 a day, so sign-up works from the first deploy. Anything else (an order confirmation, a reminder, a lead notification) isn't sent. For your own domain and wording, more volume or other email, connect your own account on Settings → Integrations → Resend, with a sender on a domain verified there; the drawer says when it isn't. See Email.

Register production's OAuth callbacks

If people sign in with Google or GitHub, the integration's drawer lists production's callback URL beside development's. Register both with the provider: one client serves both environments. Do it again after adding a domain, since the callback follows IB_APP_URL. See Google and GitHub.

Set production's own variables

Production reads development's value for every variable it has no value of its own for. Give production its own where they differ: a live Stripe key, a production Resend sender, LEADS_TO on a Site. A NEXT_PUBLIC_* value is compiled into the pages at build time, and 1Password references are resolved then, so set them before deploying rather than after. See Environment variables.

Add a domain

On Settings → Domains, add your hostname and CNAME it as the page says; make it production's primary domain and IB_APP_URL, the links in emails and the apps-domain redirect follow it. A Site's web/lib/site.ts reads its address from the same variable. Development can have a domain too. See Custom domains.

Check what roles grant

In a Members app, signed-in in accessControl/permissions.ib.ts is what the pages' scripts may read and change; unauthenticated stays empty unless a page needs it. Every members-only page, route handler and server action calls requireUser() itself. See The signed-in role.

Clear Problems

A TypeScript error in web/ fails production's build. Problems lists them in development as you type, and the deploy plan's Problems in development adds schemas that can't be read and a dev server that didn't start.

The deploy

Deploy in development's top bar, or ib-apps deploy, opens the deploy plan: what will happen, each model change with what production's records will read like, and Problems in development. Nothing changes until you choose Deploy on it. The first deploy creates production: its files, records, variables' own values, Logs and address. The build takes web/, api/ and package.json, as What a build ships says; a failed build leaves nothing live, and a later failed build leaves the live version in place.

Only in production
Schedulesschedules in settings.ib.ts run in production only, unless an entry names development too. See Schedules.
AlertsOwners and admins, and members who can see the app, get an email when production fails, and the team's webhooks an incident event; development never sends one. Who, how often and on what is set in Settings → Alerts. See Alerts.
EmailSent through your Resend, or Inference Bureau's sender for sign-in emails and invitations; never to Logs.
Preview as a userDevelopment only. Production is tested the way people use it: sign up, confirm, sign in.
VariablesRead at the deploy. A change after it waits for the next one.

After it's live

  • Open production's address and go through the flows as a visitor: sign up, the confirmation email (to an inbox now, not Logs), sign out and in, a password reset, each provider's button; on a Site, the contact form and the newsletter's confirmation.
  • Preview shows production's deployed version beside its files; Ops charts its requests by status, errors and the slowest routes from the first deploy on.
  • Production's records are its own: the user who signed up in development doesn't exist there. Fixtures a page needs are written with the API, a key with Apps Permissions, or the Playground.
  • Each later deploy ships development's files as they stand. Deploying from GitHub syncs an environment from a branch instead, and a rollback puts an earlier version back.

What to check in Logs

Look forQuery or view
Action runs that failed: an action that threw, a refused writetype = action AND state = failed
Routes answering 5xx: an unhandled error in a page or route handlerstatus >= 500
Browsers refused by CORSname ~ cors
Builds and deploys, and why one failedThe Builds and Deploys views
Background jobs out of triessource = job, or the app's Queue
Model calls and what they costcharge > 0

Emails that Resend refuses show in the app's console output, not as events: sendEmail answers { sent: false, reason: "refused" }, so a page that sends one can tell the visitor. Events are kept for the plan's retention, 1 day on Free; export what you want to keep. See Preview, Logs and Problems.

Nothing here is a deploy gate: the plan warns and the deploy goes through. The one thing that stops it is a build that fails, and Problems shows that before you start.