Skip to content

    Vibe coding

    Replit Agent to production: from workspace to a deployment that stays up

    Author

    AI Cubed

    Published

    September 9, 2026

    Read time

    8 min

    Replit Agent produces a running app inside the workspace quickly and hides most of the machinery. Production is where the machinery reappears: Deployments have their own secrets, commands and ports, and the database the agent used at one point may not be the one it's using now.

    This checklist walks the workspace-to-Deployment boundary in order and ends with a decision most people skip: whether to stay.

    The checklist, in order

    1. Real git history with a branch. Done when: a checkpoint rollback can't lose work you care about.
    2. Secrets set for the Deployment. Done when: the deployed app reads every secret it needs and no code contains a fallback value.
    3. One real database, in your account, with stranded data migrated. Done when: you can point to it and say 'this one'.
    4. Explicit build and run commands; correct host and port. Done when: the Deployment starts and serves a request.
    5. Correct deployment type. Done when: the app's behaviour (stateless, stateful, static) matches the type.
    6. Auth configured for the custom domain. Done when: every auth flow works there on a phone.
    7. Schema as migrations, run on deploy. Done when: a fresh database from migrations matches production.
    8. Cost controls on paid calls. Done when: a runaway loop would hit a limit and page you, not your card.
    9. Tests and CI. Done when: a PR that breaks checkout fails.
    10. Stay-or-move decision, written down. Done when: you know why you're where you are.
    11. Monitoring and tested backups. Done when: you've been paged by a test and restored a backup.

    Stay on Replit or move?

    If…Then…
    Launching this month, modest traffic, everything in your nameStay. Fix the Deployment and ship.
    Raising, hiring engineers, or expecting real scaleMove to a standard host (Vercel, Cloudflare, Fly) plus a managed database (Supabase, Neon). Any engineer knows the setup.
    Costs are unpredictable under loadMove, or at minimum right-size the deployment type and add limits.
    The app is a long-running process with stateReserved VM if staying; a VM or container host if moving.

    What AI Cubed does about this

    We run this checklist on Replit projects — audit in two business days including the stay-or-move recommendation, then the rescue. See finish my Replit app.

    Frequently asked questions

    Sources

    Start here

    See where your operation is losing time.

    Twenty minutes with an operator, not a salesperson. We'll name the one bottleneck costing you the most — and tell you whether it's worth fixing with software at all.

    Book your free 20-minute consult

    20 minutes · video call · no preparation needed