V0

v0

Three distinct surfaces share the name. Do not conflate them.

Getting v0 output into a repo

Four routes. Pick by how much you intend to keep.

@v0 is not a built-in shadcn namespace — it only resolves if components.json declares it. Registry mechanics, components.json, and namespace auth belong to shadcn; do not restate them here.

The Platform API also runs the seam backwards: chats.createFromRepo (GitHub URL + branch), createFromFiles (≤20 UTF-8 files), and createFromZip seed a chat from your code, so an iteration starts from what you actually shipped instead of from v0's last memory of it.

Import review checklist

v0 optimizes for a working preview, not for your repo's invariants. Walk this before the first commit; everything here is routine in generated output and cheap to fix at import time, expensive later.

Render-performance findings (waterfalls, needless re-renders, bundle weight) are react-best-practices' territory — run it after the checklist, don't duplicate it.

Keeping generated code maintainable

• Land the import as its own commit, unmodified, then fix in a second commit. The two-commit shape is what makes "what did v0 actually write" answerable later.
• Treat generated files as normal owned source the moment they merge. There is no supported re-sync of an edited file back from a chat; chats.updateFiles pushes edits into the chat, it does not pull.
• Re-generating? Feed the current repo back in with createFromRepo / createFromFiles first. Prompting a stale chat produces a diff against code you no longer have.
• Keep one chat per surface. A chat that has accumulated five unrelated screens re-emits all five.

Platform API (v2)

Verified against packages/v0-sdk/openapi.json at vercel/v0-sdk@276e1c2 (OpenAPI 3.1.2, info.version 2.0.0, server https://api.v0.dev/v2).

Auth: the default client uses V0_API_KEY, and falls back to Vercel OIDC for server-side code deployed on Vercel. createV0Client({ auth, baseUrl }) overrides both. Every call returns { data, error } — check error rather than relying on a throw.

import { v0 } from 'v0'

const res = await v0.chats.create({ message: 'Build a pricing page' })
if (res.error) throw new Error(res.error.message)

const preview = await v0.chats.getPreview({ chatId: res.data.chat.id })
// preview.data is null while the preview is still starting — poll.

const { data } = await v0.chats.getFiles({ chatId: res.data.chat.id })
// data.files: { path: 'app/page.tsx', content: string, encoding: 'utf8' | 'base64' }[]

• Streaming: createStream / sendStream return a server result — readV0Stream(result.toResponse()) yields { stream, final }.
• chats.deploy returns a deploymentId; the build runs asynchronously on Vercel — track it with the Vercel API, not by polling v0. Deploy/CI mechanics → deployments-cicd, vercel-cli.
• Webhook events: chat.created chat.updated chat.deleted message.created message.updated message.deleted message.finished.
• chats.create accepts skills (≤3) drawn from skills.sh (remote), user/team memory, or project scope.
• Never ship V0_API_KEY to a browser. @v0-sdk/react is deliberately built around app-owned proxy routes that call the server-side SDK and re-expose only what you authorize.

Models

modelConfiguration.modelId on chat/message creation, from the checked-in v2 spec: v0-mini, v0-pro (default), v0-max, v0-max-fast. Pair with imageGenerations (default false). The v1 SDK's typings also list v0-auto; it is absent from the v2 enum — read the enum in openapi.json for the version you install rather than hardcoding from here.

The Model API is a different product from the Platform API despite sharing the api.v0.dev host: an OpenAI-compatible POST /v1/chat/completions with v0-1.x-series model ids, supporting only messages, model, stream, tools, tool_choice. Reach for it when you want v0's frontend-tuned generation inside your own chat app; reach for the Platform API when you want chats, files, previews, and deploys. Routing either through AI Gateway (vercel/… provider prefix) is ai-gateway's topic — resolve ids with gateway.getAvailableModels() before hardcoding.

Verified vs not

Verified here by reading the artifact: the v2 OpenAPI document, SDK method surface, model enum, file/webhook schemas, auth model, and npm package lineage and dates. @v0 being a configured — not built-in — shadcn namespace was verified against shadcn@4.19.0's own error strings.

Not verified — v0.app, v0.dev, vercel.com, ai-sdk.dev, and api.v0.dev were all unreachable from this environment (egress policy), so the following come from secondary evidence and should be re-checked against the docs before you rely on them: the exact registry URL for the @v0 namespace; the exact "Add to Codebase" command string; the GitHub integration's branch/PR mechanics; the Model API's current model ids and whether it is still offered; and which v0 models AI Gateway carries.

References

• v0 docs — https://v0.app/docs · Platform API v2 — https://v0.app/docs/api/v2 · v1→v2 migration and AI models pages under the same tree
• vercel/v0-sdk — https://github.com/vercel/v0-sdk (read at 276e1c2; packages/v0-sdk/openapi.json is the generated SDK's copy of the upstream v2 spec)
• npm: v0 · v0-sdk · @v0-sdk/react
• Model API request body — https://v0.dev/docs/v0-model-api#request-body (cited by third-party clients; not fetched)
• In-repo: skills/vercel/shadcn (registries, components.json, theming) · skills/vercel/nextjs · skills/vercel/react-best-practices · skills/vercel/ai-gateway