To upgrade Nuxt to the latest release, use the nuxt upgrade command.
npx nuxt upgrade
yarn nuxt upgrade
pnpm nuxt upgrade
bun x nuxt upgrade
To use the latest Nuxt build and test features before their release, read about the nightly release channel guide.
latest tag is currently tracking the Nuxt v4 branch, meaning that it is particularly likely to have breaking changes right now — be careful! You can opt in to the 3.x branch nightly releases with "nuxt": "npm:nuxt-nightly@3x".Nuxt 4 is scheduled for release in Q2 2025. It will include all the features currently available through compatibilityVersion: 4.
Until the release, it is possible to test many of Nuxt 4's breaking changes from Nuxt version 3.12+.
First, upgrade Nuxt to the latest release.
Then you can set your compatibilityVersion to match Nuxt 4 behavior:
export default defineNuxtConfig({
future: {
compatibilityVersion: 4,
},
// To re-enable _all_ Nuxt v3 behavior, set the following options:
// srcDir: '.',
// dir: {
// app: 'app'
// },
// experimental: {
// scanPageMeta: 'after-resolve',
// sharedPrerenderData: false,
// compileTemplate: true,
// resetAsyncDataToUndefined: true,
// templateUtils: true,
// relativeWatchPaths: true,
// normalizeComponentNames: false,
// spaLoadingTemplateLocation: 'within',
// parseErrorData: false,
// pendingWhenIdle: true,
// alwaysRunFetchOnKeyChange: true,
// defaults: {
// useAsyncData: {
// deep: true
// }
// }
// },
// features: {
// inlineStyles: true
// },
// unhead: {
// renderSSRHeadOptions: {
// omitLineBreaks: false
// }
// }
})
When you set your compatibilityVersion to 4, defaults throughout your Nuxt configuration will change to opt in to Nuxt v4 behavior, but you can granularly re-enable Nuxt v3 behavior when testing, following the commented out lines above. Please file issues if so, so that we can address them in Nuxt or in the ecosystem.
Breaking or significant changes will be noted here along with migration steps for backward/forward compatibility.
compatibilityVersion: 4.To facilitate the upgrade process, we have collaborated with the Codemod team to automate many migration steps with some open-source codemods.
npx codemod feedback 🙏For a complete list of Nuxt 4 codemods, detailed information on each, their source, and various ways to run them, visit the Codemod Registry.
You can run all the codemods mentioned in this guide using the following codemod recipe:
# Using pinned version due to https://github.com/codemod-com/codemod/issues/1710
npx [email protected] nuxt/4/migration-recipe
# Using pinned version due to https://github.com/codemod-com/codemod/issues/1710
yarn dlx [email protected] nuxt/4/migration-recipe
# Using pinned version due to https://github.com/codemod-com/codemod/issues/1710
pnpm dlx [email protected] nuxt/4/migration-recipe
# Using pinned version due to https://github.com/codemod-com/codemod/issues/1710
bun x [email protected] nuxt/4/migration-recipe
This command will execute all codemods in sequence, with the option to deselect any that you do not wish to run. Each codemod is also listed below alongside its respective change and can be executed independently.
🚦 Impact Level: Significant
Nuxt now defaults to a new directory structure, with backwards compatibility (so if Nuxt detects you are using the old structure, such as with a top-level pages/ directory, this new structure will not apply).
srcDir is app/ by default, and most things are resolved from there.serverDir now defaults to <rootDir>/server rather than <srcDir>/serverlayers/, modules/ and public/ are resolved relative to <rootDir> by defaultcontent/ is resolved relative to <rootDir>dir.app is added, which is the directory we look for router.options.ts and spa-loading-template.html - this defaults to <srcDir>/.output/
.nuxt/
app/
assets/
components/
composables/
layouts/
middleware/
pages/
plugins/
utils/
app.config.ts
app.vue
router.options.ts
content/
layers/
modules/
node_modules/
public/
shared/
server/
api/
middleware/
plugins/
routes/
utils/
nuxt.config.ts
👉 For more details, see the PR implementing this change.
.git/ and node_modules/ folders being scanned/included by FS watchers which can significantly delay startup on non-Mac OSes.server/ and the rest of your app are running in two entirely different contexts with different global imports available, and making sure server/ isn't inside the same folder as the rest of your app is a big first step to ensuring you get good auto-completes in your IDE.app/.assets/, components/, composables/, layouts/, middleware/, pages/, plugins/ and utils/ folders under it, as well as app.vue, error.vue, app.config.ts. If you have an app/router-options.ts or app/spa-loading-template.html, these paths remain the same.nuxt.config.ts, content/, layers/, modules/, public/ and server/ folders remain outside the app/ folder, in the root of your project.tailwindcss or eslint configuration (if required - @nuxtjs/tailwindcss should automatically configure tailwindcss correctly).npx codemod@latest nuxt/4/file-structureHowever, migration is not required. If you wish to keep your current folder structure, Nuxt should auto-detect it. (If it does not, please raise an issue.) The one exception is that if you already have a custom srcDir. In this case, you should be aware that your modules/, public/ and server/ folders will be resolved from your rootDir rather than from your custom srcDir. You can override this by configuring dir.modules, dir.public and serverDir if you need to.
You can also force a v3 folder structure with the following configuration:
export default defineNuxtConfig({
// This reverts the new srcDir default from `app` back to your root directory
srcDir: '.',
// This specifies the directory prefix for `app/router.options.ts` and `app/spa-loading-template.html`
dir: {
app: 'app',
},
})
🚦 Impact Level: Moderate
Nuxt's data fetching system (useAsyncData and useFetch) has been significantly reorganized for better performance and consistency:
useAsyncData or useFetch with the same key now share the same data, error and status refs. This means that it is important that all calls with an explicit key must not have conflicting deep, transform, pick, getCachedData or default options.getCachedData: The getCachedData function is now called every time data is fetched, even if this is caused by a watcher or calling refreshNuxtData. (Previously, new data was always fetched and this function was not called in these cases.) To allow more control over when to use cached data and when to refetch, the function now receives a context object with the cause of the request.useAsyncData is unmounted, Nuxt will remove that data to avoid ever-growing memory usage.These changes have been made to improve memory usage and increase consistency with loading states across calls of useAsyncData.
// This will now trigger a warning
const { data: users1 } = useAsyncData('users', () => $fetch('/api/users'), { deep: false })
const { data: users2 } = useAsyncData('users', () => $fetch('/api/users'), { deep: true })
useAsyncData that share an explicit key (and have custom options) into their own composable:export function useUserData (userId: string) {
return useAsyncData(
`user-${userId}`,
() => fetchUser(userId),
{
deep: true,
transform: user => ({ ...user, lastAccessed: new Date() }),
},
)
}
getCachedData implementations:useAsyncData('key', fetchFunction, {
- getCachedData: (key, nuxtApp) => {
- return cachedData[key]
- }
+ getCachedData: (key, nuxtApp, ctx) => {
+ // ctx.cause - can be 'initial' | 'refresh:hook' | 'refresh:manual' | 'watch'
+
+ // Example: Don't use cache on manual refresh
+ if (ctx.cause === 'refresh:manual') return undefined
+
+ return cachedData[key]
+ }
})
Alternatively, for now, you can disable this behaviour with:
export default defineNuxtConfig({
experimental: {
granularCachedData: false,
purgeCachedData: false,
},
})
🚦 Impact Level: Minimal
The order in which modules are loaded when using Nuxt layers has been corrected. Previously, modules from the project root were loaded before modules from extended layers, which was the reverse of the expected behavior.
Now modules are loaded in the correct order:
This affects both:
modules array in nuxt.config.tsmodules/ directoryThis change ensures that:
Most projects will not need changes, as this corrects the loading order to match expected behavior.
However, if your project was relying on the previous incorrect order, you may need to:
Example of the new correct order:
// Layer: my-layer/nuxt.config.ts
export default defineNuxtConfig({
modules: ['layer-module-1', 'layer-module-2'],
})
// Project: nuxt.config.ts
export default defineNuxtConfig({
extends: ['./my-layer'],
modules: ['project-module-1', 'project-module-2'],
})
// Loading order (corrected):
// 1. layer-module-1
// 2. layer-module-2
// 3. project-module-1 (can override layer modules)
// 4. project-module-2 (can override layer modules)
If you encounter issues with module order dependencies due to needing to register a hook, consider using the modules:done hook for modules that need to call a hook. This is run after all other modules have been loaded, which means it is safe to use.
👉 See PR #31507 and issue #25719 for more details.
🚦 Impact Level: Minimal
It's possible to set some route metadata using definePageMeta, such as the name, path, and so on. Previously these were available both on the route and on route metadata (for example, route.name and route.meta.name).
Now, they are only accessible on the route object.
This is a result of enabling experimental.scanPageMeta by default, and is a performance optimization.
The migration should be straightforward:
const route = useRoute()
- console.log(route.meta.name)
+ console.log(route.name)
🚦 Impact Level: Moderate
Vue will now generate component names that match the Nuxt pattern for component naming.
By default, if you haven't set it manually, Vue will assign a component name that matches the filename of the component.
├─ components/
├─── SomeFolder/
├───── MyComponent.vue
In this case, the component name would be MyComponent, as far as Vue is concerned. If you wanted to use <KeepAlive> with it, or identify it in the Vue DevTools, you would need to use this name.
But in order to auto-import it, you would need to use SomeFolderMyComponent.
With this change, these two values will match, and Vue will generate a component name that matches the Nuxt pattern for component naming.
Ensure that you use the updated name in any tests which use findComponent from @vue/test-utils and in any <KeepAlive> which depends on the name of your component.
Alternatively, for now, you can disable this behaviour with:
export default defineNuxtConfig({
experimental: {
normalizeComponentNames: false,
},
})
🚦 Impact Level: Minimal
Unhead, used to generate <head> tags, has been updated to version 2. While mostly compatible it includes several breaking changes
for lower-level APIs.
vmid, hid, children, body.The above changes should have minimal impact on your app.
If you have issues you should verify:
useHead({
meta: [{
name: 'description',
// meta tags don't need a vmid, or a key
- vmid: 'description'
- hid: 'description'
}]
})
import { AliasSortingPlugin, TemplateParamsPlugin } from '@unhead/vue/plugins'
export default defineNuxtPlugin({
setup () {
const unhead = injectHead()
unhead.use(TemplateParamsPlugin)
unhead.use(AliasSortingPlugin)
},
})
While not required it's recommended to update any imports from @unhead/vue to #imports or nuxt/app.
-import { useHead } from '@unhead/vue'
+import { useHead } from '#imports'
If you still have issues you may revert to the v1 behavior by enabling the head.legacy config.
export default defineNuxtConfig({
unhead: {
legacy: true,
},
})
🚦 Impact Level: Minimal
When rendering a client-only page (with ssr: false), we optionally render a loading screen (from app/spa-loading-template.html), within the Nuxt app root:
<div id="__nuxt">
<!-- spa loading template -->
</div>
Now, we default to rendering the template alongside the Nuxt app root:
<div id="__nuxt"></div>
<!-- spa loading template -->
This allows the spa loading template to remain in the DOM until the Vue app suspense resolves, preventing a flash of white.
If you were targeting the spa loading template with CSS or document.queryElement you will need to update your selectors. For this purpose you can use the new app.spaLoaderTag and app.spaLoaderAttrs configuration options.
Alternatively, you can revert to the previous behaviour with:
export default defineNuxtConfig({
experimental: {
spaLoadingTemplateLocation: 'within',
},
})
error.data🚦 Impact Level: Minimal
It was possible to throw an error with a data property, but this was not parsed. Now, it is parsed and made available in the error object. Although a fix, this is technically a breaking change if you were relying on the previous behavior and parsing it manually.
Update your custom error.vue to remove any additional parsing of error.data:
<script setup lang="ts">
import type { NuxtError } from '#app'
const props = defineProps({
error: Object as () => NuxtError
})
- const data = JSON.parse(error.data)
+ const data = error.data
</script>
Alternatively, you can disable this change:
export default defineNuxtConfig({
experimental: {
parseErrorData: false,
},
})
🚦 Impact Level: Moderate
Nuxt will now only inline styles for Vue components, not global CSS.
Previously, Nuxt would inline all CSS, including global styles, and remove <link> elements to separate CSS files. Now, Nuxt will only do this for Vue components (which previously produced separate chunks of CSS). We think this is a better balance of reducing separate network requests (just as before, there will not be separate requests for individual .css files per-page or per-component on the initial load), as well as allowing caching of a single global CSS file and reducing the document download size of the initial request.
This feature is fully configurable and you can revert to the previous behavior by setting inlineStyles: true to inline global CSS as well as per-component CSS.
export default defineNuxtConfig({
features: {
inlineStyles: true,
},
})
🚦 Impact Level: Minimal
We now scan page metadata (defined in definePageMeta) after calling the pages:extend hook rather than before.
This was to allow scanning metadata for pages that users wanted to add in pages:extend. We still offer an opportunity to change or override page metadata in a new pages:resolved hook.
If you want to override page metadata, do that in pages:resolved rather than in pages:extend.
export default defineNuxtConfig({
hooks: {
- 'pages:extend'(pages) {
+ 'pages:resolved'(pages) {
const myPage = pages.find(page => page.path === '/')
myPage.meta ||= {}
myPage.meta.layout = 'overridden-layout'
}
}
})
Alternatively, you can revert to the previous behaviour with:
export default defineNuxtConfig({
experimental: {
scanPageMeta: true,
},
})
🚦 Impact Level: Medium
We enabled a previously experimental feature to share data from useAsyncData and useFetch calls, across different pages. See original PR.
This feature automatically shares payload data between pages that are prerendered. This can result in a significant performance improvement when prerendering sites that use useAsyncData or useFetch and fetch the same data in different pages.
For example, if your site requires a useFetch call for every page (for example, to get navigation data for a menu, or site settings from a CMS), this data would only be fetched once when prerendering the first page that uses it, and then cached for use when prerendering other pages.
Make sure that any unique key of your data is always resolvable to the same data. For example, if you are using useAsyncData to fetch data related to a particular page, you should provide a key that uniquely matches that data. (useFetch should do this automatically for you.)
// This would be unsafe in a dynamic page (e.g. `[slug].vue`) because the route slug makes a difference
// to the data fetched, but Nuxt can't know that because it's not reflected in the key.
const route = useRoute()
const { data } = await useAsyncData(async () => {
return await $fetch(`/api/my-page/${route.params.slug}`)
})
// Instead, you should use a key that uniquely identifies the data fetched.
const { data } = await useAsyncData(route.params.slug, async () => {
return await $fetch(`/api/my-page/${route.params.slug}`)
})
Alternatively, you can disable this feature with:
export default defineNuxtConfig({
experimental: {
sharedPrerenderData: false,
},
})
data and error values in useAsyncData and useFetch🚦 Impact Level: Minimal
data and error objects returned from useAsyncData will now default to undefined.
Previously data was initialized to null but reset in clearNuxtData to undefined. error was initialized to null. This change is to bring greater consistency.
If you were checking if data.value or error.value were null, you can update these checks to check for undefined instead.
npx codemod@latest nuxt/4/default-data-error-valueIf you encounter any issues you can revert back to the previous behavior with:
export default defineNuxtConfig({
experimental: {
defaults: {
useAsyncData: {
value: 'null',
errorValue: 'null',
},
},
},
})
Please report an issue if you are doing this, as we do not plan to keep this as configurable.
boolean values for dedupe option when calling refresh in useAsyncData and useFetch🚦 Impact Level: Minimal
Previously it was possible to pass dedupe: boolean to refresh. These were aliases of cancel (true) and defer (false).
const { refresh } = await useAsyncData(() => Promise.resolve({ message: 'Hello, Nuxt!' }))
async function refreshData () {
await refresh({ dedupe: true })
}
These aliases were removed, for greater clarity.
The issue came up when adding dedupe as an option to useAsyncData, and we removed the boolean values as they ended up being opposites.
refresh({ dedupe: false }) meant do not cancel existing requests in favour of this new one. But passing dedupe: true within the options of useAsyncData means do not make any new requests if there is an existing pending request. (See PR.)
The migration should be straightforward:
const { refresh } = await useAsyncData(async () => ({ message: 'Hello, Nuxt 3!' }))
async function refreshData () {
- await refresh({ dedupe: true })
+ await refresh({ dedupe: 'cancel' })
- await refresh({ dedupe: false })
+ await refresh({ dedupe: 'defer' })
}
npx codemod@latest nuxt/4/deprecated-dedupe-valuedata in useAsyncData and useFetch🚦 Impact Level: Minimal
If you provide a custom default value for useAsyncData, this will now be used when calling clear or clearNuxtData and it will be reset to its default value rather than simply unset.
Often users set an appropriately empty value, such as an empty array, to avoid the need to check for null/undefined when iterating over it. This should be respected when resetting/clearing the data.
If you encounter any issues you can revert back to the previous behavior, for now, with:
export default defineNuxtConfig({
experimental: {
resetAsyncDataToUndefined: true,
},
})
Please report an issue if you are doing so, as we do not plan to keep this as configurable.
pending value in useAsyncData and useFetch🚦 Impact Level: Medium
The pending object returned from useAsyncData, useFetch, useLazyAsyncData and useLazyFetch is now a computed property that is true only when status is also pending.
Now, when immediate: false is passed, pending will be false until the first request is made. This is a change from the previous behavior, where pending was always true until the first request was made.
This aligns the meaning of pending with the status property, which is also pending when the request is in progress.
If you rely on the pending property, ensure that your logic accounts for the new behavior where pending will only be true when the status is also pending.
<template>
- <div v-if="!pending">
+ <div v-if="status === 'success'">
<p>Data: {{ data }}</p>
</div>
<div v-else>
<p>Loading...</p>
</div>
</template>
<script setup lang="ts">
const { data, pending, execute, status } = await useAsyncData(() => fetch('/api/data'), {
immediate: false
})
onMounted(() => execute())
</script>
Alternatively, you can temporarily revert to the previous behavior with:
export default defineNuxtConfig({
experimental: {
pendingWhenIdle: true,
},
})
useAsyncData and useFetch🚦 Impact Level: Medium
When using reactive keys in useAsyncData or useFetch, Nuxt automatically refetches data when the key changes. When immediate: false is set, useAsyncData will only fetch data when the key changes if the data has already been fetched once.
Previously, useFetch had slightly different behavior. It would always fetch data when the key changed.
Now, useFetch and useAsyncData behave consistently - by only fetch data when the key changes if the data has already been fetched once.
This ensures consistent behavior between useAsyncData and useFetch, and prevents unexpected fetches. If you have set immediate: false, then you must call refresh or execute or data will never be fetched in useFetch or useAsyncData.
This change should generally improve the expected behavior, but if you were expecting changing the key or options of a non-immediate useFetch, you now will need to trigger it manually the first time.
const id = ref('123')
const { data, execute } = await useFetch('/api/test', {
query: { id },
immediate: false
)
+ watch(id, () => execute(), { once: true })
To opt out of this behavior:
// Or globally in your Nuxt config
export default defineNuxtConfig({
experimental: {
alwaysRunFetchOnKeyChange: true,
},
})
useAsyncData and useFetch🚦 Impact Level: Minimal
The data object returned from useAsyncData, useFetch, useLazyAsyncData and useLazyFetch is now a shallowRef rather than a ref.
When new data is fetched, anything depending on data will still be reactive because the entire object is replaced. But if your code changes a property within that data structure, this will not trigger any reactivity in your app.
This brings a significant performance improvement for deeply nested objects and arrays because Vue does not need to watch every single property/array for modification. In most cases, data should also be immutable.
In most cases, no migration steps are required, but if you rely on the reactivity of the data object then you have two options:
- const { data } = useFetch('/api/test')
+ const { data } = useFetch('/api/test', { deep: true })
export default defineNuxtConfig({
experimental: {
defaults: {
useAsyncData: {
deep: true,
},
},
},
})
npx codemod@latest nuxt/4/shallow-function-reactivitybuilder:watch🚦 Impact Level: Minimal
The Nuxt builder:watch hook now emits a path which is absolute rather than relative to your project srcDir.
This allows us to support watching paths which are outside your srcDir, and offers better support for layers and other more complex patterns.
We have already proactively migrated the public Nuxt modules which we are aware use this hook. See issue #25339.
However, if you are a module author using the builder:watch hook and wishing to remain backwards/forwards compatible, you can use the following code to ensure that your code works the same in both Nuxt v3 and Nuxt v4:
+ import { relative, resolve } from 'node:fs'
// ...
nuxt.hook('builder:watch', async (event, path) => {
+ path = relative(nuxt.options.srcDir, resolve(nuxt.options.srcDir, path))
// ...
})
npx codemod@latest nuxt/4/absolute-watch-pathwindow.__NUXT__ objectWe are removing the global window.__NUXT__ object after the app finishes hydration.
This opens the way to multi-app patterns (#21635) and enables us to focus on a single way to access Nuxt app data - useNuxtApp().
The data is still available, but can be accessed with useNuxtApp().payload:
- console.log(window.__NUXT__)
+ console.log(useNuxtApp().payload)
🚦 Impact Level: Medium
Child folders in your middleware/ folder are also scanned for index files and these are now also registered as middleware in your project.
Nuxt scans a number of folders automatically, including middleware/ and plugins/.
Child folders in your plugins/ folder are scanned for index files and we wanted to make this behavior consistent between scanned directories.
Probably no migration is necessary but if you wish to revert to previous behavior you can add a hook to filter out these middleware:
export default defineNuxtConfig({
hooks: {
'app:resolve' (app) {
app.middleware = app.middleware.filter(mw => !/\/index\.[^/]+$/.test(mw.path))
},
},
})
🚦 Impact Level: Minimal
Previously, Nuxt used lodash/template to compile templates located on the file system using the .ejs file format/syntax.
In addition, we provided some template utilities (serialize, importName, importSources) which could be used for code-generation within these templates, which are now being removed.
In Nuxt v3 we moved to a 'virtual' syntax with a getContents() function which is much more flexible and performant.
In addition, lodash/template has had a succession of security issues. These do not really apply to Nuxt projects because it is being used at build-time, not runtime, and by trusted code. However, they still appear in security audits. Moreover, lodash is a hefty dependency and is unused by most projects.
Finally, providing code serialization functions directly within Nuxt is not ideal. Instead, we maintain projects like unjs/knitwork which can be dependencies of your project, and where security issues can be reported/resolved directly without requiring an upgrade of Nuxt itself.
We have raised PRs to update modules using EJS syntax, but if you need to do this yourself, you have three backwards/forwards-compatible alternatives:
getContents().es-toolkit/compat (a drop-in replacement for lodash template), as a dependency of your project rather than Nuxt:+ import { readFileSync } from 'node:fs'
+ import { template } from 'es-toolkit/compat'
// ...
addTemplate({
fileName: 'appinsights-vue.js'
options: { /* some options */ },
- src: resolver.resolve('./runtime/plugin.ejs'),
+ getContents({ options }) {
+ const contents = readFileSync(resolver.resolve('./runtime/plugin.ejs'), 'utf-8')
+ return template(contents)({ options })
+ },
})
Finally, if you are using the template utilities (serialize, importName, importSources), you can replace them as follows with utilities from knitwork:
import { genDynamicImport, genImport, genSafeVariableName } from 'knitwork'
const serialize = (data: any) => JSON.stringify(data, null, 2).replace(/"\{(.+)\}"(?=,?$)/gm, r => JSON.parse(r).replace(/^\{(.*)\}$/, '$1'))
const importSources = (sources: string | string[], { lazy = false } = {}) => {
return toArray(sources).map((src) => {
if (lazy) {
return `const ${genSafeVariableName(src)} = ${genDynamicImport(src, { comment: `webpackChunkName: ${JSON.stringify(src)}` })}`
}
return genImport(src, genSafeVariableName(src))
}).join('\n')
}
const importName = genSafeVariableName
npx codemod@latest nuxt/4/template-compilation-changes🚦 Impact Level: Minimal
compilerOptions.noUncheckedIndexedAccess is now true instead of false.
This change is a follow up to a prior 3.12 config update where we improved our defaults, mostly adhering to TotalTypeScript's recommendations.
There are two approaches:
nuxt.config.ts:export default defineNuxtConfig({
typescript: {
tsConfig: {
compilerOptions: {
noUncheckedIndexedAccess: false,
},
},
},
})
🚦 Impact Level: Minimal
Nuxt now generates separate TypeScript configurations for different contexts to provide better type-checking experiences:
.nuxt/tsconfig.app.json - For your app code (Vue components, composables, etc.).nuxt/tsconfig.server.json - For your server-side code (Nitro/server directory).nuxt/tsconfig.node.json - For your build-time code (modules, nuxt.config.ts, etc.).nuxt/tsconfig.shared.json - For code shared between app and server contexts (like types and non-environment specific utilities).nuxt/tsconfig.json - Legacy configuration for backward compatibility.nuxt/tsconfig.json will continue to work as before.typescript.nodeTsConfig option: You can now customize the TypeScript configuration for Node.js build-time code.This change provides several benefits:
For example, auto-imports are not available in your nuxt.config.ts (but previously this was not flagged by TypeScript). And while IDEs recognized the separate context hinted by tsconfig.json in your server/ directory, this was not reflected in type-checking (requiring a separate step).
No migration is required - existing projects will continue to work as before.
However, to take advantage of improved type checking, you can opt in to the new project references approach:
tsconfig.json to use project references:{
"files": [],
"references": [
{ "path": "./.nuxt/tsconfig.app.json" },
{ "path": "./.nuxt/tsconfig.server.json" },
{ "path": "./.nuxt/tsconfig.shared.json" },
{ "path": "./.nuxt/tsconfig.node.json" }
]
}
tsconfig.json files (like server/tsconfig.json) that extended .nuxt/tsconfig.server.json.- "typecheck": "nuxt prepare && vue-tsc --noEmit"
+ "typecheck": "nuxt prepare && vue-tsc -b --noEmit"
app/ directory.server/ directory.shared/ directory.app/, server/, or shared/ directories will not work with the new project references setup.export default defineNuxtConfig({
typescript: {
// Customize app/server TypeScript config
tsConfig: {
compilerOptions: {
strict: true,
},
},
// Customize build-time TypeScript config
nodeTsConfig: {
compilerOptions: {
strict: true,
},
},
},
})
The new configuration provides better type safety and IntelliSense for projects that opt in, while maintaining full backward compatibility for existing setups.
🚦 Impact Level: Minimal
Four experimental features are no longer configurable in Nuxt 4:
experimental.treeshakeClientOnly will be true (default since v3.0)experimental.configSchema will be true (default since v3.3)experimental.polyfillVueUseHead will be false (default since v3.4)experimental.respectNoSSRHeader will be false (default since v3.4)vite.devBundler is no longer configurable - it will use vite-node by defaultThese options have been set to their current values for some time and we do not have a reason to believe that they need to remain configurable.
polyfillVueUseHead is implementable in user-land with this pluginrespectNoSSRHeaderis implementable in user-land with server middlewaregenerate Configuration🚦 Impact Level: Minimal
The top-level generate configuration option is no longer available in Nuxt 4. This includes all of its properties:
generate.exclude - for excluding routes from prerenderinggenerate.routes - for specifying routes to prerenderThe top level generate configuration was a holdover from Nuxt 2. We've supported nitro.prerender for a while now, and it is the preferred way to configure prerendering in Nuxt 3+.
Replace generate configuration with the corresponding nitro.prerender options:
export default defineNuxtConfig({
- generate: {
- exclude: ['/admin', '/private'],
- routes: ['/sitemap.xml', '/robots.txt']
- }
+ nitro: {
+ prerender: {
+ ignore: ['/admin', '/private'],
+ routes: ['/sitemap.xml', '/robots.txt']
+ }
+ }
})
In the table below, there is a quick comparison between 3 versions of Nuxt:
| Feature / Version | Nuxt 2 | Nuxt Bridge | Nuxt 3+ |
|---|---|---|---|
| Vue | 2 | 2 | 3 |
| Stability | 😊 Stable | 😊 Stable | 😊 Stable |
| Performance | 🏎 Fast | ✈️ Faster | 🚀 Fastest |
| Nitro Engine | ❌ | ✅ | ✅ |
| ESM support | 🌙 Partial | 👍 Better | ✅ |
| TypeScript | ☑️ Opt-in | 🚧 Partial | ✅ |
| Composition API | ❌ | 🚧 Partial | ✅ |
| Options API | ✅ | ✅ | ✅ |
| Components Auto Import | ✅ | ✅ | ✅ |
<script setup> syntax | ❌ | 🚧 Partial | ✅ |
| Auto Imports | ❌ | ✅ | ✅ |
| webpack | 4 | 4 | 5 |
| Vite | ⚠️ Partial | 🚧 Partial | ✅ |
| Nuxt CLI | ❌ Old | ✅ nuxt | ✅ nuxt |
| Static sites | ✅ | ✅ | ✅ |
The migration guide provides a step-by-step comparison of Nuxt 2 features to Nuxt 3+ features and guidance to adapt your current application.
If you prefer to progressively migrate your Nuxt 2 application to Nuxt 3, you can use Nuxt Bridge. Nuxt Bridge is a compatibility layer that allows you to use Nuxt 3+ features in Nuxt 2 with an opt-in mechanism.