Documentation
Build an app, review what changed, ship it to your own account
NoCoder writes a working app from a description, then shows you every file it changed before that change lands. This page covers the whole loop: the modes you can build in, how review works, which models you can run it on, and how deploying and exporting your code work.
Quickstart
You need an account and nothing else. No credit card, and no API key unless you want to use your own.
- 1
Describe what you want built
Open the builder and write a plain description of the app. Be specific about the things you would notice if they were wrong: who logs in, what gets saved, what the main screen shows. A one-line prompt gets a one-line app. - 2
Pick the app mode
Landing page, single-page app, or full-stack. This decides whether you get a database and API routes at all, so set it before the first run. See choosing an app mode. - 3
Watch it work, and read the diffs
The agent plans, writes files, installs packages and runs the build. Every file it touches shows up as a diff you can open. - 4
Open the preview
A dev server runs inside your workspace and the preview pane points at it. This is the real app rather than a mockup, so click through it. - 5
Keep going, or ship it
Ask for changes in the same conversation, then deploy it or download the code.
Choosing an app mode
Mode is the most consequential setting on the page, because it decides how much of an app gets built. Full-stack for a calculator wastes a run; single-page app for something that needs logins produces an app that looks finished and saves nothing.
Landing
One static marketing page. No routing, no backend.
Hero, features, pricing, testimonials, footer.
SPA
A multi-route React app with client-side state. No backend.
Calculators, dashboards on mock data, portfolio sites.
Full-stack
Frontend, a Prisma database, API routes and real auth.
Anything with accounts, saved data, or more than one user.
If you describe something that needs a backend while the mode is set to landing or single-page app, NoCoder says so and offers to switch rather than quietly building the wrong thing. Sign-in that only stores a name in the browser is the specific failure that check exists to prevent.
Full-stack plans before it writes: entities first, then API routes, then screens. The schema is decided once, rather than drifting as files get written.
Reviewing changes
Review is the part that makes everything else safe. Turn on Review changes and nothing the agent writes is applied until you approve it. Each file arrives as a diff, and you accept or reject them one at a time.
With review off, changes apply as they are made and you read the diffs afterwards. Either way every change is recorded, so you can see what was touched and roll back to an earlier point.
Review is worth the extra clicks on anything you intend to keep. Reading a diff costs seconds. Discovering in week two that you cannot tell what changed is the reason this product exists.
The workspace
Each project gets its own workspace: a container running a full VS Code (code-server) environment, with the file tree, an editor and a terminal. It is a real machine rather than a preview of one. You can edit files by hand, run commands and install packages, and the agent sees the same filesystem you do.
The dev server runs there too, and the preview pane points at it. When the agent runs a build or a test it reads the actual exit code, so a failing build is treated as a failure instead of something it has to infer from the log text.
Free-plan workspaces are put to sleep when idle and woken on your next visit. Your files are kept.
Models and your own key
Google Gemini, Ollama and NVIDIA NIM models are available on every plan at no cost. Paid models are included on paid plans within a fixed monthly budget, so spend cannot run away.
You can bring your own Ollama on any plan, including Free. Connect your own Ollama (local or Ollama Cloud) and NoCoder sends requests to it, so you pay your provider directly for tokens and pay us nothing for them. The plan buys the workspace, not the model.
Models fail differently from one another. If a build stalls, or a model keeps producing the same broken file, switching models is usually faster than re-prompting the same one.
Deploying
Deploying puts the app on a public URL on Vercel. NoCoder creates the project, sends the build, clears the SSO protection Vercel applies to new projects so the app is actually reachable, and reports the result back into the editor.
Deploying to your own Vercel account
By default, deploys go out through NoCoder's shared Vercel account. You can connect your own instead, from the deploy dialog. After that, every app you deploy is created as a project in your Vercel account: your plan, your domains, your billing. You own the project and the deployment, and the app keeps running whether or not you keep using NoCoder.
Disconnecting stops future deploys going to your account and touches nothing already deployed. If you remove the integration from Vercel's side instead, NoCoder is notified and drops the stored token, so it never tries to deploy with a credential you have revoked.
Full-stack deploys
A full-stack deploy also provisions a Postgres database and wires the connection string into the deployment. Redeploys reuse that same database so your data survives, and schema changes go through a generated migration that is inspected first: if a change would drop a table or a column, the deploy stops and names the statement it refused to run, rather than discarding data quietly.
Exporting your code
You can download the whole project as a zip at any time, on any plan. It is ordinary source code, the same files you have been reading in the workspace, with no NoCoder runtime or wrapper around them. Run it locally with the usual install and dev commands for the stack.
This is deliberate. An app you cannot take with you is not an app you own, and an export you cannot run is not an export.
Plans and limits
The Free plan includes a workspace, storage and the free-tier models, with a daily cap on runs that use the shared paid models. Bringing your own Ollama takes token cost out of the picture entirely.
Paid plans raise the workspace, storage and request limits, and add a monthly budget for paid models. The current numbers live on the pricing page, which is the source of truth; this page deliberately does not restate them.
Troubleshooting
The preview is blank
Usually the dev server is still starting, or the build failed. Check the build output in the workspace terminal. Asking the agent to fix the build error directly is faster than describing the blank screen to it.
The app looks unstyled
The CSS pipeline did not start. Ask the agent to check the Tailwind and PostCSS configuration for the project's framework. This is a config file, not a missing stylesheet.
Sign-in does not persist
Almost always a single-page app being asked to do full-stack work: there is no database, so the session has nowhere to live. Switch the mode to full-stack and ask for real authentication.
The agent stopped early
Say what is still missing, in the same conversation. It keeps the project context, so “the delete button does nothing” is a more useful next message than restating the original prompt.
Getting help
Email [email protected] with the project name and what you expected to happen. If a build failed, the error from the workspace terminal is the single most useful thing you can include.
Longer write-ups on how the agent works, including where it fails, are on the blog.
Build something and read the diffs
Start free with Gemini and Ollama models, or bring your own Ollama. No credit card required.