Skip to main content
How to host a vibe-coded app: from prototype to production
DENEVERSE team

How to host a vibe-coded app: from prototype to production

Your Lovable or Cursor app works on the first try. Production needs more: safe keys, a backed-up database, and a server someone actually looks after.

Written by DENEVERSE team, Managed infrastructure specialists.

The most reliable way to take a vibe-coded app to production is this: move the code into a Git repository, package it in a Docker container, and deploy it to a virtual server (VPS) or Kubernetes. Run the database separately and back it up off the server. Before any of that, review security. AI writes working code fast; it does not write secure code as reliably.

Why hosting suddenly becomes the problem

Vibe coding means you don't write the app line by line; you describe it to an AI tool such as Lovable, Bolt, v0, Replit, Cursor, or Claude Code. By the end of the afternoon you have a working prototype. The trouble starts when you want real customers to use it.

The hosting built into these tools is great for demos. In production you hit limits: credits that drain with every visit, little control over where data lives, no separation between test and production, or free plans that take the site offline until the next billing cycle once you exceed a limit.

Security first, hosting second

In 2025, Veracode tested more than 100 language models on 80 coding tasks. In 45% of cases the model chose an implementation with a known OWASP Top 10 vulnerability, and newer or larger models were not meaningfully better. AI-written code isn't bad; it just hasn't been reviewed.

Before you go public, check at least the following:

  • Secret keys don't belong in code or in the frontend. Any variable prefixed with NEXT_PUBLIC_ or VITE_ is visible to every visitor in the browser.
  • Treat any key that ever appeared in the repository or in an AI chat as leaked, and generate a new one.
  • On Supabase, enable Row Level Security on every table and make sure the policies don't give every signed-in user access to all data. The simplest test: sign in as a second user and try to load the first user's data.
  • Enforce permissions on the server, not just by hiding a button in the UI.
  • Rate-limit sign-in, sign-up, and expensive AI API calls.
  • Turn off debug mode and verbose error messages in production.

Step 1: Get the code out of the tool

Cursor and Claude Code work directly in your local folder, so you already have the code. Lovable syncs the project to GitHub, and Bolt and Replit can export to GitHub or as a ZIP. The goal is always the same: code in a Git repository that you own.

Git isn't a formality. Every deployment becomes a specific version you can roll back to with one command, and .gitignore keeps the .env file with your passwords out of the repository and off the server.

Step 2: Find out what kind of app you actually have

  • A static frontend, typically React with Vite from Lovable: the build produces a folder of files that any web server can serve.
  • An app with a server side (Next.js, Nuxt, SvelteKit, Express, FastAPI): it needs a running process, which means a virtual server or a container.
  • Backend as a service (Supabase, Firebase): the database, sign-in, and file storage run elsewhere, and you need to decide whether they stay there.

Step 3: Package the app in a Docker container

Docker is the most practical route for vibe-coded apps. A Dockerfile describes how the app is built and started. A docker-compose.yml file then connects the app, the database, and a reverse proxy with automatic HTTPS. AI can write both files for you; just have someone who understands operations review them.

The big advantage: the same container runs on your laptop, on a virtual server, and in a Kubernetes cluster. When the app grows, you don't have to rewrite anything.

Step 4: Choose where the app will run

  • Docker VPS: a virtual server with Docker and a web interface such as Portainer. The best first production step for one app and a small team.
  • Linux VPS: full control of the system over SSH when you need something outside containers.
  • Managed Kubernetes: when you have several services, separate test and production environments, and want zero-downtime deployments through GitOps.

Step 5: Database, backups, and monitoring

The most valuable part of your app is neither the code nor the server; it's the data. AI can write the code again. It can't recreate your users' data.

  • Run the database separately from the app, ideally as a managed service.
  • Back up automatically and off the app server. Restore a backup now and then, or you won't know whether it works.
  • When leaving Lovable Cloud or Supabase, expect that secrets are not exported and must be set again, and that users will need to sign in again after the move.
  • Set up monitoring and alerts so you learn about an outage before your customers do.

Step 6: Separate development from production

An AI tool is happy to "just quickly tweak" things you never asked it to touch. Keep at least two environments: a test environment for new versions and a production environment that only receives verified code. Deploying straight from the Git repository through CI/CD or GitOps ensures nothing reaches production without review.

When to stay on the AI tool's own hosting

For a prototype, an internal tool for a few colleagues, or testing an idea with early users, the built-in hosting in Lovable or Bolt is fine. Moving makes sense once the app processes personal data, earns money, or has more users than your credits can cover.

Summary

Vibe coding has shrunk the path from idea to prototype to hours. The path from prototype to production is still the same: reviewed code in Git, a container, a separate and backed-up database, monitoring, and two environments. Follow it and you have an app you can rely on.

Don't want to run the server yourself? At vibecode.deneverse.cz we deploy and operate your vibe-coded app for you, on DENEVERSE infrastructure in Prague and Bratislava.

vibecode.deneverse.cz

For vibe-coded apps, DENEVERSE offers Docker VPS with Docker Compose and Portainer, Linux VPS, managed PostgreSQL, MongoDB, and MariaDB, and managed Kubernetes with Argo CD for larger projects. Everything runs on our own infrastructure in Prague and Bratislava. Tell us what you've built and we'll talk through the path to production.

Need a stable platform for your applications?

Contact us

More articles