The Bundler Under Your Lovable App Was Replaced. An Eight-Point Vite 8 Check.
Lovable can now hand you a ZIP of your codebase. What that ZIP builds with is a separate question - in March 2026 Vite replaced its bundler wholesale. Here is the check we run before the code leaves the platform.
The export button is new. The toolchain it exports into is not the one you started on.
Lovable's 23 September 2026 changelog added downloadable project codebases: a ZIP from Git settings, on paid plans. That is a genuine ownership improvement, and it is the first thing founders ask us about once they start thinking about leaving. The code is yours and now you can hold it.
What the ZIP does not tell you is which build toolchain that code expects. For a React and Vite project, that question got considerably more interesting this year.
Vite 8 shipped on 12 March 2026, and it did not add a bundler. It replaced the ones that were already there. esbuild and Rollup are out; Rolldown and Oxc are in, a single Rust pipeline in place of two JavaScript ones, which is where the release's 10-30x faster builds figure comes from. Vite 8.1 followed on 23 June 2026 with an experimental bundled dev mode. While your project lives inside Lovable's managed build, none of this is your problem. The day you run the build yourself, all of it is.
Here is the check we run before a Lovable codebase moves onto a client's own infrastructure. It is eight points and it is deliberately boring.
1. Read which Vite major the project actually expects
Open package.json in the export and read the vite dependency before you run a single command. Everything below branches on that answer. Do not infer it from when the project was created, and do not take anyone's word for it. Read the file.
2. Check your Node floor before anything else
Vite 8 requires Node.js 20.19+ or 22.12+, the same floor as Vite 7, set so the toolchain can ship ESM-only. If your CI image or your deploy target is older, this fails first, before any of the more interesting problems get a chance to surface. The fix is usually one line in a Dockerfile, which is exactly why it is worth finding on day one rather than during a deploy.
3. Rename the config keys that moved
Vite 8 ships a compatibility layer that auto-converts existing esbuild and rollupOptions configuration to its Rolldown equivalents, so many projects need no config changes at all in order to build. The deprecated names still work. That is convenient, and it is also how a config quietly drifts out of date. If you are taking ownership of this repository, rename them now, while you have the context - not in eighteen months when nobody remembers why there is an esbuild block in a project with no esbuild in it.
| Vite 7 configuration | Vite 8 equivalent | What changed underneath |
|---|---|---|
build.rollupOptions | build.rolldownOptions | Bundling runs through Rolldown |
optimizeDeps.esbuildOptions | optimizeDeps.rolldownOptions | Dependency pre-bundling runs through Rolldown |
esbuild config block | oxc | JavaScript transforms run through Oxc |
build.minify: 'esbuild' | Oxc minifier | A different minifier, and different output |
| CSS minification | Lightning CSS by default | A different CSS engine, and different edge cases |
Every row in that table is a deprecation rather than a break. That is the point. Nothing fails, and the behaviour underneath has changed anyway.
4. Re-test your CommonJS default imports
This is the item most likely to bite silently. In Vite 8 the default import from a CommonJS module is handled consistently between dev and build, where the two could previously disagree. The migration guide says plainly that this may break existing code.
If any dependency in your prototype is CommonJS and you import its default export, exercise that code path in a real build, not just in dev, before you trust the output. There is an escape hatch, legacy.inconsistentCjsInterop: true, but it is documented as deprecated. Use it to ship this week if you must. Do not use it as a resting place.
5. Look at the browser target you just inherited
Vite 8 raises the default build.target to Chrome 111, Edge 111, Firefox 114 and Safari 16.4, roughly the features that landed two and a half years ago. For most founder apps that is entirely fine and no user will notice.
But it is a product decision, and a version bump just made it on your behalf. If you have a stated minimum browser, an enterprise customer on a managed fleet, or a market where older Safari is common, set build.target explicitly rather than inheriting whatever the next major decides.
6. Account for what was removed, not changed
These fail loudly, which makes them the good kind of problem. You find them in the first build rather than in production.
- Output formats
systemandamdare no longer supported. import.meta.urlis no longer polyfilled in UMD and IIFE builds.- Passing a URL to
import.meta.hot.accepthas been removed. build.rollupOptions.watch.chokidarhas been removed.- Decorator lowering is not supported. If your code uses decorators, you are adding Babel or SWC back into a toolchain whose appeal was not needing them.
7. Audit the plugins one at a time
Most Vite plugins work under Rolldown. Most is doing real work in that sentence, and a prototype assembled quickly tends to accumulate plugins nobody has audited since.
Three specifics from the migration guide. Plugins that transform non-JS content must now declare moduleType: 'js'. Several Rollup hooks are unsupported: shouldTransformCachedModule, resolveImportMeta and renderDynamicImport. And build() now throws a typed BundleError carrying an .errors array, so any custom build script that parsed error strings needs rewriting. Count your plugins, then check them individually. This is the step people skip and then spend a day on.
8. Budget the install, and take the staged path
Vite 8's install is around 15 MB larger than Vite 7: roughly 10 MB of Lightning CSS and 5 MB of Rolldown binary. On a laptop that is nothing. In a container image rebuilt on every deploy it is a line item worth knowing about before it arrives as build minutes.
Then the sequencing the Vite team themselves recommend for complex projects: run the rolldown-vite package on Vite 7 first, confirm the build is clean, and only then take the major. It turns one risky step into two boring ones, which is the trade you want on a codebase you did not hand-write.
What this check is not
This is a check on the build, not on the whole move. It says nothing about your database, your secrets or your hosting. One item in particular does not appear in the ZIP at all: Lovable documents on-request prerendering for older React and Vite apps as a behaviour of deployed public URLs, served only to verified search and AI crawlers. That is a property of the hosting, not a file in your repository, so it does not travel with the export. We priced the rest of that bill separately.
Where we land on it
On a small prototype this list is an afternoon. On a real app with a plugin pipeline and a CI image it is a week, and most of that week goes on finding out which of the eight points apply rather than on fixing them.
If you would rather not spend it, our Strategy Audit is $499 one-time, credited toward any build, and it ends with a written technical roadmap rather than a proposal. A Publishable MVP is $4,999. The Founding Client build is $2,999 and carries a 14-day build guarantee with a full refund if we miss the deadline. Every build carries the same revenue guarantee: first paying customers within 90 days of launch, or full refund.
Or book the free 20-minute intro call and tell us what your package.json says.
Sources
- Announcing Vite 8 - https://vite.dev/blog/announcing-vite8
- Announcing Vite 8.1 - https://vite.dev/blog/announcing-vite8-1
- Vite migration guide - https://vite.dev/guide/migration
- Lovable changelog - https://docs.lovable.dev/changelog
- Lovable SEO documentation - https://docs.lovable.dev/features/seo