For v0
v0 can look finished. The preview link asks people to sign in first.
A locked v0 preview link is a security setting working as intended, not a broken app. Its auto-fix patches code errors as you go, it doesn't try your app the way a customer would. If something's wrong after v0 generates your app, the place to look is the GitHub repo v0 already keeps up to date.
Last verified 18 August 2026. Model facts are Vercel's own engineering blog, not our review of v0 quality.
-
Before it writes anything
It works from context, within limits
Your instructions, the chat history, relevant docs, examples, and anything you uploaded. It's working from what you've given it, not reading your mind.
-
Writing the code
A licensed AI model, not v0's own invention
Vercel has said different versions of v0 use different underlying Claude models. The point is: v0 is a wrapper around a rented model, and that model can be swapped out.
-
While it's generating
An auto-fix pass, not a test
A separate step trained to catch and patch common code mistakes as they happen. That is not the same as someone clicking through your app.
-
Your preview link
Asks for sign-in by default
New projects lock preview links behind a sign-in automatically. Getting around that needs a secret only you, the owner, have. PeerRun will never try to pull that secret from a link you paste in.
What v0 actually is
A few tools working together, not one AI that just "gets" web apps.
The v0 model family combines specialized knowledge from retrieval-augmented generation, reasoning from state-of-the-art large language models, and error fixing from a custom streaming post-processing model.
In plain terms: v0 takes a general-purpose AI model, feeds it extra reference material about how to build web apps well, and runs a separate cleanup step to catch common code mistakes. That combination is why a v0 preview can look so polished. It is not a knock on Vercel's engineering, and it does not tell you anything about whether your specific app works.
Small edits and large changes are handled by different versions of the underlying model, and quality is checked with internal tests plus that auto-fix step, both aimed at the code, not at whether a real person can use the app. v0 can browse the web while generating, but only to look at reference images and design examples, not to check that your app actually works. We found no v0 product that tests user journeys.
The technical detailThe v0 model family combines retrieval-augmented generation, a rented frontier large language model (Vercel has said v0-1.0-md used Claude Sonnet 3.7 and v0-1.5-md used Sonnet 4), and a trained post-processor called AutoFix (vercel-autofixer-01) for common generated-code errors.
Cited: v0 composite model family, agentic features. PROMPT_TO_APP_PIPELINES Part 1 and 2; TARGET_ACCESS §3B.2.
This is working exactly as designed
A preview link asking for sign-in is not a broken app.
New v0 projects lock preview links behind a sign-in by default, as a security setting. Only a link you deliberately publish to production is meant to be public. Getting around that lock needs a special secret that only you, the project owner, have.
PeerRun will not try to get around that sign-in. If you paste a locked preview link, the honest answer is that we could not get in. That is Unknown, not Failed. Bring us the GitHub repo instead, or a production link you actually intend to be public.
The technical detailDeployment Protection with Vercel Authentication is enabled by default for new projects. Bypass requires an owner-supplied x-vercel-protection-bypass header, optionally with a cookie or query parameter.
Cited: Vercel changelog, Vercel Deployment Protection docs. TARGET_ACCESS §3.1, §3.3.
How to bring a v0 app
Once GitHub is connected, that repo is your real code.
Once you connect a GitHub repository, v0 saves your code there automatically as you chat, and that repo becomes your project's real, current version. v0 does not keep a separate copy anywhere else.
That is what makes this simple: sign in to PeerRun and bring that same repository. We do not have a special v0 integration, and we have not tested a large, statistically meaningful sample of v0 apps.
The technical detailv0 creates a working branch per chat (v0/username-abc123), auto-commits every code-changing message, and opens pull requests against a base branch.
Cited: v0 GitHub docs. TARGET_ACCESS §5.7.
Honesty
What PeerRun will not be able to tell you
- A locked preview link
- A protected preview stays closed to us. We will not try to pull a bypass secret out of your account. That's Unknown or Blocked, not Failed.
- Signed-in product
- Anything behind your sign-in stays Unknown until you grant access. We can see the login page exists. We cannot see what's behind it.
- A bot wall
- A robot-check is Blocked, not broken. We have no way around it.
- What we do not claim
- We do not get around Vercel's sign-in wall. We do not say v0 "has no quality." Its auto-fix is real, at the code level. We do not have a measured failure rate for v0 apps.
Questions people actually ask
v0, specifically
My v0 app looks broken in the preview. Is that auto-fix failing?
Auto-fix catches common code mistakes as your app is generated. A broken user journey can still exist underneath code that otherwise compiles fine. We will not diagnose your specific chat from this page. Bring us the repo if you want the app actually tried.
Can I just paste my v0 chat link?
A chat or preview link is usually locked behind sign-in. That is the same wall described above. The GitHub branch is what we can actually work with.
Is my production link on Vercel enough on its own?
A public production link is reachable, and anyone can open it. But it only proves the logged-out parts work unless you grant us more access. The repo remains the deeper way in.
Sources
What this page is allowed to say
- Introducing the v0 composite model family
- v0 GitHub docs, agentic features
- Vercel Deployment Protection changelog and docs. In-repo: TARGET_ACCESS §3.1, §3.3, §5.7; PROMPT_TO_APP_PIPELINES Part 1 and 2
A pretty v0 preview means the tools did their job. It is not a verdict on your app.
Sign in and bring the GitHub repository v0 already writes to.
We will not invent a pass. Anything behind your sign-in stays Unknown until you grant access. A robot-check is Blocked, not broken. The first look is on us.