cloudparts

v0

Vercel’s prompt-to-app agent: it writes a Next.js app inside a Vercel Sandbox VM, pushes each chat’s changes to its own GitHub branch, and publishes to Vercel.

What it is
AI app builders
Type
Commercial product
Made by
Vercel Inc.
License
Proprietary
Facts checked
13 October 2026

What it is

You describe an app in a chat, and v0’s agent plans the work, writes the code, runs it and shows a live preview next to the conversation. It defaults to Next.js with Tailwind CSS and shadcn/ui, because server actions and API routes let one codebase hold both the UI and the backend. Since the rebuild in February 2026, each chat runs in its own Vercel Sandbox, a lightweight VM with Node.js, pnpm, npm, yarn and bun, a dev server that detects Next.js or Vite, and a terminal you can type into alongside the agent. The VM’s files persist between visits to the same chat; a pooled sandbox starts with a 30-minute lifetime, active ones are extended 10 minutes at a time, and none lives longer than 24 hours.

v0 is tied to Vercel and GitHub. Databases, blob storage and AI providers come from the Vercel Marketplace (Supabase, Neon, Upstash, fal, Grok and others), apps call models through Vercel’s AI Gateway by default, and publishing creates or updates a Vercel project. Connect a GitHub repository and v0 creates a working branch per chat, commits every change there, gives the branch its own preview deployment, and on **Publish** opens and merges a pull request, respecting branch protection and required checks. You can also import an existing repository, so the agent works on a real codebase instead of a throwaway demo.

Billing is metered in tokens against a credit balance. Every prompt also sends chat history, source files and Vercel-specific context as input tokens, so short prompts can still cost a lot. The Free plan allows 7 messages a day; Plus is $30 per user a month with $30 of credits each, and Business is $100 per user a month and keeps your data out of training by default. v0’s own models are priced from $0.20 per million input tokens (v0 Mini) to $5 (v0 Max) and $10 (v0 Max Fast), and there is a TypeScript SDK and REST API for embedding the same agent in other products.

Why teams pick it

  • Code lands in your own GitHub repository through branches, preview deployments and pull requests, so engineers can review what a product manager or designer built
  • The preview runs in a real Node.js VM, so API routes, server code and database connections behave as they will on Vercel
  • Strong at React and Next.js UI: shadcn/ui and Tailwind by default, design systems, Figma import and a visual Design Mode
  • A v2 API and SDK expose the same agent, sandbox previews and deployment to other apps and agents

Watch out for

  • Built around Vercel: deployment, integrations and billing all go through a Vercel project and the Vercel Marketplace
  • Next.js gives the most reliable results; other frameworks work less predictably
  • Token billing includes the whole chat context, so long chats on large codebases use credits quickly; generation pauses when credits run out
  • Business plan ($100 per user a month) or above is needed to keep your data out of model training by default

Alternatives

What all ai app builders have in common →Compare side by side →

How it fits with other technologies

Create an app from a prompt with the v0 API (v2) · ts

import { v0 } from 'v0'

const result = await v0.chats.create({
  message: 'Create a project management dashboard with a kanban board',
})

if (result.error) {
  throw new Error(result.error.message)
}

console.log('Created chat:', result.data.chat.id)

See it in a real stack

No architecture uses v0 yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2023–2026

Vercel introduced v0 in 2023 as “Generative UI”: describe an interface, get React code built on open-source tools (React, Tailwind CSS and shadcn/ui), pick one of the iterations, and copy and paste it into your own app. The goal was to help developers build the first version of a product. After 100,000 people joined the waitlist in three weeks, Vercel moved it from alpha to beta on 11 October 2023 and introduced paid plans, charging credits per generation.

v0 became generally available in 2024. On 11 August 2025 Vercel renamed v0.dev to v0.app and made it agentic: instead of one model call per prompt, an agent plans the work, searches the web, inspects sites, reads files, checks its own output and adds integrations, aimed at people who don’t write code as well as developers.

Prototypes that lived outside real codebases kept needing rewrites, so on 3 February 2026 Vercel rebuilt v0 around a sandbox runtime. It can import any GitHub repository, pulls environment variables from Vercel, and gives every chat a git branch, pull requests and deploy-on-merge. Vercel said more than 4 million people had used v0 by then.

Why it’s built this way

Output real code on open-source components
From the start v0 generated React with Tailwind CSS and shadcn/ui, so the result could be pasted into an existing app and developed further rather than locked in a builder.
An agent instead of a single generation
With v0.dev, getting a good result often meant prompting again for fixes, design or features. The agentic v0.app plans, checks its work and fixes errors itself, to need fewer prompts.
A VM sandbox and git branches instead of a browser preview
The old browser-based preview couldn’t run server code, API routes or real database connections, and prototypes outside the real codebase had to be rewritten. A Vercel Sandbox VM behaves like the deployed app, and git branches let non-engineers ship through pull requests.

Timeline

  1. 2023

    Announced as Generative UI; beta and paid plans on 11 October after a 100,000-person waitlist

  2. 2024

    Generally available

  3. 2025

    v0.dev becomes v0.app, an agent, on 11 August

  4. 2026

    The new v0 on 3 February: sandbox runtime, GitHub import, branches and pull requests

Sources14 · facts checked 13 October 2026

Improve this page on GitHub