Why it matters

Claude Code can build your app on its own. So why build it on NodeGX?

We asked Claude Code for the same app twice, once on its own and once through NodeGX. The two first versions looked the same. What differs is what you can still do with the app once people use it and start asking for changes.

What you get

Three things you can still do later

The three are in order of how much we can show you today.

Open it

An editor for your app, without a rewrite

A NodeGX app is a graph of parts, saved in a format a program can read. When a client wants to change their own prices, your agent can build them an editor for just that. The editor reads and writes the same files as the app.

Claude Code has built a 3D editor, a children’s block editor, a Bubble-style editor and a map of a backend this way, each in under a day. In an app written as plain code, an editor has to be cut into the code. The further the code has gone, the more that costs.

See the editors →
Start loaded

Your assistant starts with the method

We’ve built templates, games and test apps with four AI coding tools, and kept a record of where they went wrong. The same few things come up: no plan, no play-test, and a bug fixed once and then forgotten.

npm create nodegx sets your folder up so the first conversation starts the right way. Your assistant scopes the app with you and writes the brief and the plan into the project before it builds. More of what we’ve learnt goes in next.

Start from npm →
Leave

Choose React or Swift when you know you need it

You don’t have to pick a framework on the first day and find out with real users that it was the wrong one. Build on the graph, then export when you know what you need.

Export to React writes an app with Vite and TypeScript. Every node in the picker exports. Before it writes anything, it tells you what won’t come across. Swift is planned after React.

Code export is in alpha. Explore the code and build on it, but don’t ship a production app from it yet.

Already have an app in code?

We converted a coded game to see what it costs

Claude Code converted a JavaScript game into a NodeGX app in 70 minutes, for $34.90. Every screen came across as NodeGX parts. A few features were lost on the way. The page lists them.

See both versions →

Why it holds

What keeps the blueprint safe to build on

Restraint

The agent builds from known parts

Code the agent writes outside the nodes, in script nodes, is counted. The editor shows that number for the whole app and for each component. An app quietly turning into JavaScript is visible.

Testing

Every node has a written spec

Every implementation of a node is tested against its spec in our continuous integration. An agent that uses a Query Records node gets the behaviour the spec describes, not whatever it remembered.

Security

One permission model, in a file

The backend has four kinds of rule: public, signed in, a named role or nobody. The rules are a file in your project. A security report a person can read is built and waiting for a release.

Who this is for

Four kinds of project

Learning

People learning to build

A graph shows where the data comes from and what happens when a button is pressed. A wall of generated code doesn’t. We built a maths game for children, a coding game where children program robots with blocks, and a guide for people who have never built an app. The games were play-tested with children before they went further.

Budget

Projects on a small budget

Where an AI coding subscription is the whole budget, NodeGX adds no fee. The editor and the packages are free and open source. We have measured the same apps built on Mistral’s subscription and on cheaper models through the Cline and OpenCode tools, with the same checks, so a builder can choose by price.

Ownership

Projects the builder must own

Your project is a folder of files on your disk. Put it in git, open it in any editor, hand it to Claude Code, or move it to a server you rent. Nothing in it needs an account with us.

Architecture

Projects where the architecture matters

A client asks where their data lives. They ask who can see it. They ask why the app does one thing when they click another. With a NodeGX project the person who built it answers from the graph, not from a chat transcript. The backend’s rules are a file in the project. A map of them is one click away.

What we are not saying

NodeGX is not the fastest way to a small app

Asking Claude Code for a React app is quicker than asking it for a NodeGX app. For a small tool you’ll use for a week, plain code is fine. NodeGX is for the app you’ll still be answering questions about in a year.

We measured this. One brief, built by Claude Code alone and by Claude Code with NodeGX, gave two first versions that looked the same, and the judges found nothing a person needs at that size. The next test is the one a client sets a month in: an editor for the settings, a page only the teacher can see, a changed rule. That number goes here when we have it.