Meet Nuxi: the Nuxt agent gets a name

On 9 June the Nuxt team published Meet Nuxi. The AI agent on nuxt.com, in beta since April, gets a name, Nuxi, and three new features: GitHub sign-in with saved conversations, branching chats and sharing them. Longer-standing Nuxt users know the name from the terminal: for a while, nuxi dev was the command that started a dev server.

If you only use the CLI, nothing changes. It still answers to both nuxi and nuxt, and nuxt is the form the Nuxt 4.4 docs use for every command. The commands a learner types most, from nuxt dev to nuxt info, are further down in that form.

What Nuxi is meant to do

The April beta already knew the docs, searched GitHub issues and showed modules and templates as cards. The 9 June post gives the agent an identity to match the Nuxt brand. The team says the old command name has always meant a lot to them, so they are bringing it back as the agent's name.

The post lists four aims for the agent:

• Playful. It reacts to where you are and what you do, and a few details are left for you to find. The post gives two hints: type hi, or hover over the Nuxi icon in the header and move your mouse around.
• Context-aware. It knows which page you are reading, so asking "how do I use this in my app?" on a docs page needs no further explanation.
• Grounded. Answers come through the Nuxt MCP server, which reads the official docs, the modules catalogue, templates and changelogs. When you ask about an error, it searches issues in the nuxt, nuxt-modules and nuxt-content organisations before it answers.
• Built to grow. Modules, templates, hosting providers and blog posts appear as clickable cards, a StackBlitz playground can open straight from the conversation, and a feedback form is one click away.

Sign in, branch and share chats

The team calls these the biggest batch of changes since the April beta:

• Sign in with GitHub at /login, and your conversations are saved. A question started on your phone can be finished on your laptop.
• Branch a chat. Under any answer, Branch in new chat copies the conversation up to that point into a fresh thread and leaves the original untouched. Use it to try a second approach without losing the first.
• Share an answer. Conversations are private by default. You can mark one public and share the link. Anyone can read it, nobody else can post in it, and you can switch it back.

Limits, model and privacy

The FAQ at the end of the post answers the practical questions:

• You do not need an account. Without one you lose the history across devices, and your messages are counted per IP address instead of per account.
• Twenty messages a day is the limit for now. The team says it is still tuning that number.
• The model is Anthropic's Claude Sonnet 4.6, reached through the Vercel AI Gateway, with the Nuxt MCP server as its source.
• You can delete a conversation from /dashboard/chat.
• What it knows about you when you are signed in: your GitHub username, your avatar and your conversations on nuxt.com, and nothing about your repositories or code.

Nuxi the agent, nuxi the package, nuxt the command

Three names now sit close together.

Nuxi, with a capital N, is the agent. You talk to it in the browser.

nuxi is an npm package, the original home of the Nuxt CLI (command-line interface: the program you run in a terminal). The same code is also published as @nuxt/cli, which the nuxt package depends on, so every Nuxt project already has it. Both packages register one program under four command names: nuxi, nuxt, nuxt-cli and nuxi-ng.

nuxt is the name the documentation uses for every command, and the one a new project's scripts call.

So npx nuxi dev and npx nuxt dev run exactly the same code, and old tutorials that say nuxi still work when you paste them. No CLI release came with the 9 June post. The most recent one was a patch on 11 May that stopped a needless restart of the dev server on a cold start.

Write nuxt anyway. It matches the docs the agent answers from, and now that Nuxi names an assistant, a README that says "run nuxi" reads like an invitation to open a chat. A new project's package.json already does this:

{
  "scripts": {
    "build": "nuxt build",
    "dev": "nuxt dev",
    "generate": "nuxt generate",
    "preview": "nuxt preview",
    "postinstall": "nuxt prepare"
  }
}

The postinstall line runs nuxt prepare after every install. It creates the .nuxt folder and generates the types your editor needs, so autocompletion works before you have started the dev server once.

Create, run and build a project

These commands take a project from an empty folder to a deployable build. The docs write them with npx; inside a project, npm run dev and the other scripts call the same commands.

npm create nuxt@latest my-app
cd my-app
npx nuxt dev --open
npx nuxt build
npx nuxt preview
npx nuxt generate

npm create nuxt@latest runs the create-nuxt package. It asks which template to use and downloads it into my-app. Then it asks for a package manager and whether to start a Git repository, installs the dependencies, and offers to browse modules.

Flags answer questions up front: --packageManager pnpm picks the package manager, and --no-modules skips the module prompt. With npm, put them after --, or npm reads them as its own options: npm create nuxt@latest my-app -- --packageManager pnpm --no-modules.

nuxt dev starts the development server on http://localhost:3000 with hot module replacement, which swaps changed code into the running page without a full reload. --open opens the browser and --port picks another port. --host with no value listens on every network interface, so a phone on the same network can reach the server.

nuxt build writes a .output folder holding the server and everything it needs in production.

nuxt preview, also available as nuxt start, runs that build on your machine. It loads your .env file for convenience; production will not load it for you, so set real environment variables on the host.

nuxt generate pre-renders every route to plain HTML for static hosting. It is nuxt build with prerendering switched on.

Add modules, upgrade Nuxt and check a project

You reach for the next four once a week rather than once a minute.

npx nuxt module add pinia
npx nuxt upgrade --dedupe
npx nuxt typecheck
npx nuxt info

nuxt module add installs a module with your package manager, adds it to package.json and puts it in the modules array of nuxt.config.ts, so there is nothing to edit by hand. List several names to add them in one go. To look a module up first, run npx nuxt module search pinia; it only shows modules compatible with your Nuxt version.

nuxt upgrade moves Nuxt to the latest stable release. Without a flag it asks what to do with your lockfile: dedupe it, recreate it along with node_modules, or skip. --dedupe picks the first answer up front, so duplicate dependencies are tidied after the upgrade. There is also a --channel option for nightly builds; leave it alone in an app you ship.

nuxt typecheck regenerates the types and runs vue-tsc over the whole app. It uses typescript and vue-tsc from the project when both are installed. Without them it runs both through npx every time, which is slower, so install them once with npm install -D vue-tsc typescript.

nuxt info prints your operating system, Node version, CLI version, package manager, Nuxt and Nitro versions, builder, config keys and modules. It also copies the same details to your clipboard as a Markdown table ready for a GitHub issue.

When to ask Nuxi and when to run --help

Nuxi is someone you ask; nuxt is something you run.

Use --help for flags. npx nuxt dev --help prints every option of the CLI you have installed, offline and with no daily limit. The docs and the agent describe the latest release, which is usually what you have, but not always.

Use Nuxi when you do not know which command or module you need, or when an error message means nothing to you. Open it on the docs page you are stuck on, so it starts with the right context.