browser · harness · product

v0

An app-building agent that turns a prompt or repo into a working web app with a live preview and one-click deploy, now with a GA headless API.

watchevidence: mixedby Vercelv0.app ↗

v0

  • Maker: Vercel
  • URL: https://v0.app ; API https://api.v0.dev/v2
  • Status (10 October 2026): commercial product, closed source. Launched in 2023 as a UI generator; now an app-building agent that edits files and runs a dev server in a Vercel Sandbox. The v0 API (v2) became generally available on 5 August 2026 per a third-party write-up. Pricing is credit-based (Free with $5/month, Plus $30, Business $100 per user). Vercel’s June 2026 Ship event added agent infrastructure around it.

What it is

An agent that turns a prompt (or a repo, a ZIP, a screenshot) into a working web app, with a live preview URL and one-click deploy to Vercel. Historically it generated React with Tailwind and shadcn/ui; we couldn’t confirm the current framework from the API announcement, though Vercel’s ecosystem makes Next.js the likely default.

The problem it’s solving

Building a real front end is slow. v0 makes the first version of an app nearly free and keeps it on a path to production (Git, deploys, a real framework).

Its path / bet

The durable end of the HTML camp: the model writes real application code that people keep, deploy and maintain. It’s in this cluster because it’s the reference point for “the agent writes the UI as code”, not because it does throwaway screens.

How it works

  • chats.create / createAsync / createStream run the agent; the stream emits text, thinking, file edits, bash commands and tool calls.
  • The agent works in an isolated sandbox, runs a dev server and checks its work against it.
  • createVercelProject and deploy ship it. Preview tokens are short-lived and should be proxied.

Strengths

  • Production-grade output on a mainstream stack; the generated code is the asset.
  • A headless API, so other agents and products can embed the generator.
  • The agent verifies against a running app, which is a form of read-back.

Weaknesses / limits

  • Heavy for a ten-minute screen: a sandbox, a dev server, a framework, seconds to minutes per turn.
  • Cloud and Vercel bound; no terminal surface; credits cost money.
  • The agent writes code, so every UI is a project to maintain.

Relation to fictty

Watch. Barely overlaps. It’s useful as the far “durable” pole on the ephemerality axis: it shows what the HTML camp looks like when the UI is meant to last, and therefore what fictty is not.

Could fictty adopt it instead of building?

No. Different problem.

What fictty should take from it

  • Stream the agent’s work as typed parts (text, file edits, commands). fictty’s watch output and history should be typed events too, not free text.
  • Graduation path. If a fictty screen turns out to be worth keeping, the honest next step is a real app; a “describe this screen as a spec for v0 or a ratatui app” export could be cheap and useful. Low priority.

Sources

Not verified: the GA date against Vercel’s own changelog, and the current default framework.