Vercel's v0 API v2 Lets AI Agents Build and Ship Apps Without Humans

v0 API v2 is now in public beta, and it goes well beyond what the original Platform API offered. v1 gave you a REST endpoint that turned prompts into code files. v2 gives you the full agent: a VM-backed chat that can start from a prompt, a GitHub repo, or a ZIP archive, render a live dev server preview, accept follow-up messages, and deploy straight to Vercel, all programmatically.

From chat interface to headless app builder

The key shift in v2 is that the underlying runtime changed entirely. Vercel had already moved v0 toward production readiness with a sandbox-based runtime that mirrors real environments. That brought native GitHub integration for branches and PRs, support for API routes, server actions, and database connections, and the ability to import and work on existing codebases. The API v2 exposes all of that programmatically.

The ability to start v0 chats from GitHub repos via API signals that Vercel wants v0 embedded into CI/CD pipelines and internal tooling, not just used interactively.

What changed from v1

The v1 API organized work around projects and versions. v2 drops that model entirely. In v2, the chat is the durable workspace and messages are its history. You no longer juggle chat IDs, project IDs, and version IDs for common workflows. Most operations come down to acting on the chat itself or sending it messages.

v2 chats also run on VMs rather than the browser-based next-lite runtime from v1. VMs support a fuller project environment, richer file operations, and a more direct path to Vercel projects, previews, and deployments.

The execution model is now explicit too. Rather than a single endpoint with synchronous or asynchronous behavior controlled by flags, v2 splits those modes into distinct endpoint variants:

• Sync endpoints block until the result is complete
• Async endpoints return immediately and let you poll the message or chat
• Streaming endpoints return Server-Sent Events so your UI can render progress as the agent thinks, edits files, uses tools, and reports usage

The core API surface

The SDK installs from npm and the base URL is https://v0.app/api/v2. Here's the minimal flow from prompt to deployed app:

import { v0 } from 'v0'

// 1. Start a chat (or use createFromZip / createFromRepository)
const chat = await v0.chats.create({
  message: 'Build a dashboard with charts',
})

// 2. Iterate with follow-up messages
await v0.messages.sendStream({
  chatId: chat.id,
  message: 'Add dark mode support',
})

// 3. Deploy to Vercel
const deployment = await v0.chats.deploy({
  chatId: chat.id,
})

Install the SDK (currently on the canary channel) with:

npm install v0@canary

Authentication reads from process.env.V0_API_KEY by default. Generate a key at v0.app settings.

What you can build with it

Developers are already using the API to power website builders where users describe a site and get production-ready code, Slack and Discord bots that return deployed web apps, VSCode plugins and CLIs built around the prompt-to-app workflow, and embedded flows in analytics or CRM tools that generate UI components from natural language.

The v2 docs outline five main use-case categories:

• Custom chat interfaces — embed v0's generation loop inside your own UI using @v0-sdk/react
• Automated workflows — trigger code generation from CI/CD events, PRs, or background jobs using async endpoints
• AI agents — connect v0 as a tool via the v0 MCP server, so any MCP-compatible agent can spin up and deploy apps
• Internal tooling — generate and deploy dashboards or admin panels from structured data or specs
• Educational platforms — interactive coding tutorials where each step produces a live, runnable preview

Integrations through MCP

v0's third-party integrations run through MCP: Stripe, Supabase, Neon, Shopify, Snowflake, Linear, Notion, and Glean. By betting on the protocol layer rather than building bespoke connectors, v0's integration surface grows as the MCP ecosystem grows. You can register MCP servers against a chat via the API, so generated apps can call live services during generation.

Strengths and limitations

Generation quality for React and Next.js frontend work is strong. Complex backend logic still needs to be wired in separately, so v0 works best for UI components, landing pages, and dashboards rather than full-stack applications with heavy authentication or database logic.

Vercel ecosystem lock-in is worth considering. One-click deployment, GitHub sync, and environment variable import all work seamlessly, but only within Vercel's infrastructure. If your deployment target is elsewhere, you can pull generated files via Download Chat Files and deploy independently, but you lose the native integration.

Pricing and access

API access requires at minimum the Premium plan. Each plan comes with a monthly credit pool that draws down every time you generate a component, chat with the agent, or call the API.

Who this is actually for

v0 has grown from a UI component generator into a full-stack development environment with git integration, a terminal, merge conflict resolution, and enterprise features. The API v2 extends that trajectory into team and enterprise territory. Signed commits, SAML SSO, Snowflake integration, and Platform API v2 are features aimed at teams with compliance requirements, data governance needs, and existing infrastructure.

The broader unlock is composability. Any system that can make an HTTP call can now trigger a full app-generation cycle, receive a live preview URL, and push to production without a human in the loop. That's the foundation for AI agents that don't just write code, but ship it.