# Lovable app to production

A Lovable app is production ready when secrets, a tested backup, and rollback are in place. What publish and a Git export leave behind.

Canonical URL: https://unbroker.cloud/blog/ai-built-app-to-production
Published: 2026-10-29
Reviewed: 2026-10-29
Author: Unbroker

## The short answer

An AI-built app is ready for other people's data when the secrets are outside the code, a backup has been restored once, a broken deploy can be reverted, and you find out if it falls over.

A page that loads is the prototype. Lovable's own publish already hosts a snapshot. Moving that app, or one from Bolt or Cursor, onto your account does not create the backup or the rollback for you. The five questions below are the gap. Setup steps for Unbroker stay on [Deploy a Lovable app](https://unbroker.cloud/deploy/lovable) and the [production checklist for AI-built apps](https://unbroker.cloud/deploy/cursor).

## It works is not the same as in production

A prototype works when you can open it. It is in production when a bad deploy, a leaked key, or a deleted row has a path back, and when that path has been tried.

Real users bring data that is not yours to lose. Lovable documents that you own the code and the data, and that leaving its managed hosting means you take on databases, storage, backups, and access rules. That handoff is the point of this list, not a claim that the builder failed.

## 1. Where are the secrets?

Keys in the frontend bundle or in a committed .env ship with the code. Anyone who can read the repository or the built files can read them.

Generated apps often start that way because the key was in the chat or in a local `.env` that got synced. Lovable's external-hosting notes say the values you need can sit in that `.env` and in the synced repository. Treat that file as a list of names, then move the values.

<aside class="plan-box">
<p><strong>On Unbroker.</strong> Put values on the service. Each variable is scoped All, Production, or Preview, and a scoped value overrides All for that environment. A preview copies All and Preview, never Production-only values. Pull request previews require Pro. The variable screen is part of the <a href="/deploy/cursor">Cursor checklist</a>.</p>
</aside>

## 2. What data survives a redeploy?

The container filesystem is temporary. A new deploy can replace the container and leave anything written only inside it behind.

A database is for rows you need to query. A volume is for files the app expects on a local path. Object storage is for files you want outside the container. If the database is still Lovable's or Supabase's, its access rules stay there. [Row level security](https://supabase.com/docs/guides/database/postgres/row-level-security) is a Postgres setting on that project: a table in an exposed schema without it is readable and writable by any role that has a grant. Enabling it is not something the host does for you.

<aside class="plan-box">
<p><strong>On Unbroker.</strong> A persistent volume survives a redeploy, a restart, and a rollback of the code. Reset, detach, and deleting the service remove the volume's data. Volumes require Pro, and a service with a volume runs one replica. Managed Postgres also requires Pro. The smallest catalogue plan is 1 vCPU, 1 GB of RAM, and 10 GB of storage at $21.21 per month, separate from the Pro fee of $25 per month with $25 of usage credit. Free does not include volumes, managed databases, or Dockerfiles. What a $0 plan can still run, with no payment method, is <a href="/blog/free-backend-hosting">Free backend hosting</a>. Details are on <a href="/solutions/persistent-files">persistent storage</a> and <a href="/managed-databases/postgres">managed Postgres</a>.</p>
</aside>

## 3. Do you have a backup you have restored?

A copy you have never restored is a hope. The backup has to land somewhere you control, on a schedule, with a retention count, and you have to bring one back before launch.

Git sync is a copy of the code, not of the rows or the uploaded files. Lovable says to export the database and download storage files separately, then check them at the destination before you retire the original.

<aside class="plan-box">
<p><strong>On Unbroker.</strong> Volume backups go to an S3-compatible bucket you add, with your credentials. You set a schedule and how many to keep, or run one by hand, and the history lists each run. The documented example is <code>ubctl apps backups schedule svc_123 --destination dst_123 --cron "0 3 * * *" --retain 7</code>. Backups require Pro. For a managed cluster, availability and recovery depend on the plan you select, and the docs say to test a restore before you rely on the store. The Postgres health card shows version, size, and connections. It reads statistics, not your rows.</p>
</aside>

## 4. What does a rollback undo?

A rollback puts an earlier deployment back. On Unbroker it promotes that deployment and does not rebuild the commit. It does not put the database or the volume back to an older copy.

The volume file that was written after the bad release is still there. A row that was updated is still updated. That is why backup and rollback are two plans. Catching the bad release before merge is a third one: a staging environment, or a pull request preview, runs the change away from production.

<aside class="plan-box">
<p><strong>On Unbroker.</strong> <code>ubctl apps rollback</code> promotes a previous deployment. An extra environment starts stopped, and Promote here deploys the images you tested, without a rebuild. GitHub pull request previews require Pro, and closing the pull request removes that preview. The command and the preview rule are on the <a href="/deploy/cursor">Cursor page</a>.</p>
</aside>

## 5. How do you know when it breaks?

If nobody is watching, a failed release looks like a quiet product. You want a health check that refuses a bad revision, logs from the running service, and a message when a deploy finishes or spending crosses a line you set.

Lovable tells you to check status.lovable.dev and the Publish dialog when a published site is down. That covers Lovable's hosting. It does not cover a copy you run somewhere else.

<aside class="plan-box">
<p><strong>On Unbroker.</strong> Advanced health checks are a Pro feature. A failed health check keeps an unhealthy revision from being treated as live, and runtime logs stay on the service. Deploy results can go to email or a Slack webhook, for successes, failures, or both. Cost alerts take a monthly budget and thresholds such as 50%, 80%, and 100%, plus a forecast alert. Those controls are in the <a href="/changelog/2026-06-24-deploy-notifications-and-cost-alerts">notifications release</a>.</p>
</aside>

## Moving an exported app to your own host

The export is the code. It is not the database rows, and it is not the files in storage.

Lovable's Git sync, including github.com, is on all plans and is two-way. A one-time zip of the codebase needs a paid plan. Apps created from May 13, 2026 use TanStack Start and run server code. Older apps are React and Vite and build to static files. Shareable preview links are work in progress and expire after 7 days by default. Publishing itself puts a snapshot on a public URL with HTTPS, and that hosting is part of Lovable. The walkthrough for deploying the repository on Unbroker is [Deploy a Lovable app](https://unbroker.cloud/deploy/lovable). It is not repeated here.

Bolt's help center says you can [connect GitHub](https://support.bolt.new/integrations/git) so the project is backed up outside Bolt, publish with Bolt hosting, the Netlify integration, or another service, and [download the project as a zip](https://support.bolt.new/building/using-bolt/projects-files) from Export. A new Bolt repository starts private on `main`. Bolt commits a change that does not break the project, and it checks GitHub every 30 seconds for outside edits. Bolt does not document, on the pages checked for this post, whether that zip includes database rows. Do not assume it does. If the tool already pushed the repo, the Unbroker path is the same repository flow as the [Cursor checklist](https://unbroker.cloud/deploy/cursor).

## Pre-launch checklist

Do these before you send real users. Each line is one check.

1. Move secret values out of the frontend bundle and out of the repository.
2. Set production values in the host, and keep them off previews.
3. Name where the rows live, and where the uploaded files live.
4. If the database is Supabase, turn on row level security there and test it there.
5. Treat the container filesystem as temporary.
6. Attach a volume, or a database, for anything that must survive a redeploy.
7. Send backups to a destination you control, with a schedule and a retention count.
8. Restore one backup before launch.
9. Confirm a rollback brings code back and leaves the database and volume where they are.
10. Run the change in a preview or a staging environment before production.
11. Turn on a health check, read the runtime logs once, and set a deploy notification.
12. Set a spend alert if the host has one.

## FAQ

### Is a Lovable app production ready?

Not because the page loads. Lovable's publish action puts a snapshot on a public HTTPS URL, and that is hosting. Production also means secrets are out of the repo, a backup has been restored once, a bad deploy can be reverted, and you hear about a failure.

### How do I take a Bolt or Lovable app to production?

Get the code into a Git repository you control, then decide who runs the database, the files, and the backups. Lovable documents Git sync on all plans. Bolt documents a GitHub connection and a zip download. The Unbroker steps for that repository are on the linked pages, not repeated here.

### Where should I store API keys in an AI-built app?

In the host's service variables, not in the frontend bundle and not in a committed .env. On Unbroker each variable is All, Production, or Preview, and a preview does not receive Production-only values.

### Does a rollback restore my database?

No. On Unbroker, ubctl apps rollback promotes an earlier deployment and does not rebuild that commit. It does not restore a database or a volume. Those need their own backup and a restore you have already tried.

### How do I back up an app built with Lovable?

If the database and files still live in Lovable, follow Lovable's own export for records and storage files. Git sync does not include them. If you move the data to Unbroker, a volume backup goes to an S3-compatible bucket you control, and that feature requires Pro.

### What does a Lovable export to GitHub include?

The application source, including database migration files that describe the structure. It does not include the rows in the database or the files in storage. A one-time zip download is a paid-plan option. Git sync itself is on all plans.

### What should I check before launching a vibe-coded app?

Secrets, what survives a redeploy, one tested restore, what a rollback does not undo, and how you hear about a failure. The checklist at the end of this post is that list.

### Do I need a staging environment for a small app?

You need a place to try the change before real users see it. On Unbroker, an extra environment starts stopped, and Promote here deploys the images you already tested. A pull request preview is the lighter version, and it requires Pro.

## Related reading

- [Deploy a Lovable app](https://unbroker.cloud/deploy/lovable)
- [Production checklist for AI-built apps](https://unbroker.cloud/deploy/cursor)
- [Persistent storage and volume backups](https://unbroker.cloud/solutions/persistent-files)
- [Managed Postgres](https://unbroker.cloud/managed-databases/postgres)
- [Letting a coding agent deploy safely](https://unbroker.cloud/blog/coding-agent-deploy-safely)
- [Pricing](https://unbroker.cloud/pricing)
