AI Builder (Puter)

Why it's ranked here

The case is the engineering under the hood, not the pitch. The test command chains roughly a hundred standalone test scripts covering publish races, rollback, multi-tab saves and preview trust, and a production build refuses to run until they pass. Template checks fail the build rather than a user's project. The chat index code recovers from corrupt or missing files instead of silently wiping history. It is Apache 2.0 per the README and runs from a clone with two commands. The limit is that every useful feature routes through one vendor's platform.

What's good

Data safety gets real attention. When the saved project list is unreadable, the app rebuilds it from the individual project files and blocks further writes if even that fails, so a network blip cannot erase your sidebar. Two open tabs merge their changes rather than overwriting each other. Each official template is validated at build time: a size cap of 5 MB, a 2,000 character limit on its extra model instructions, and every backend placeholder must name a backend the template actually ships. The sample chat template sanitizes model output before rendering it.

Tradeoffs

This is a front end for Puter, not a standalone builder. Sign-in, storage, AI calls, backend functions and hosting all go through Puter's SDK, which loads from Puter's servers even when you run the builder yourself, so self-hosting does not remove the dependency. Apps it generates load a small script from the builder's own site that adds a Made with Puter badge and the click-to-edit hook. The interface sends usage data to Plausible analytics. The code is plain browser scripts sharing one global scope, which makes it harder to fork safely than a modular codebase.

How to use it well

Use it when you want a small site or app online fast and are happy for it to live on Puter: a landing page, a booking page for a local business, a simple AI chat tool. The templates lean toward exactly that, with local service businesses and small AI apps. It is a poor fit if you need to own your hosting, choose your own database, or hand code to a team on a conventional stack. If you fork it, run the full test suite before each build, since the build depends on it anyway.

Observed

License

Apache 2.0, as stated in README.md

Language

JavaScript, plain browser scripts bundled by Vite (vite.config.js)

Install

git clone, npm install, npm run dev (README.md)

Platform dependency

Puter SDK loaded from Puter's CDN for auth, storage, AI and hosting (src/index.html)

Interfaces

Browser app; connects to remote MCP servers via @modelcontextprotocol/sdk (package.json)

Tests

Standalone Node test scripts chained in the test command, run before every build (package.json)

Template validation

Build fails on templates over 5 MB or with unmatched worker placeholders (scripts/build-templates.mjs)

Analytics

Plausible script in the builder page (src/index.html)

Default model

claude-opus-5-5 set in src/js/app.js