Carrd

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me a one-page site instead of paying for Carrd: a profile, a launch page or a link hub, hosted on my own domain, with a form that works.

STACK
- Astro with a Node adapter, or static output if there is no form
- Vanilla CSS and a little vanilla JS. No framework — this is one page
- SQLite through better-sqlite3 only if the form or the analytics need it
- Caddy in front, with automatic TLS

THE PAGE
- One HTML document, server-rendered or built. Under 100KB including the font, and rendering complete in under a second on a phone over a poor connection
- Sections as components: hero, about, links, features, gallery, testimonials, pricing, form, footer. Each takes its content from one data file, so changing the words never means editing markup
- Content in a single `site.js` or a Markdown file — the whole point of the product being replaced is editing without touching code, and a data file is that
- Responsive from 320px to a wide desktop with no horizontal scrolling anywhere
- Dark and light, following the system and toggleable, with the choice stored

TYPE AND COLOUR
- One self-hosted variable font as WOFF2, subset to the characters actually used. Never a font from a third-party CDN: it is slower, it breaks when they do, and in the EU it is a documented privacy problem
- A small token set: two backgrounds, two text colours, one accent, one border. Every colour pair measured against WCAG AA rather than chosen by eye
- One size scale used everywhere, and one spacing scale

THE FORM
- Fields defined in the data file; rendered, validated on the server, and stored
- A honeypot field plus a minimum time-on-form, which stops nearly all of it, and a rate limit per salted address hash
- No third-party form service, no captcha that sends your visitors to somebody else's server
- Submissions stored in SQLite and emailed on, with the email being a convenience and the database being the record
- A real success state that replaces the form, and an error state that keeps what was typed
- Works with JavaScript disabled: a plain POST, a redirect, a message

LINKS, IF IT IS A LINK HUB
- A list from the data file, each with a label, an icon and a URL
- Clicks counted through your own redirect, so you know which link people use — first-party, no third party involved
- Icons as inline SVG from the file, not an icon font and not a script

WHAT MOST ONE-PAGE SITES GET WRONG
- Open Graph and Twitter tags, with an image generated at build time from the site's own title and colours using satori and resvg
- A favicon set from one source image, including the sizes iOS and Android actually use
- JSON-LD: Person or Organization, with sameAs pointing at every profile
- A sitemap, a robots.txt, and a canonical URL. Yes, for one page
- Semantic HTML: one h1, real landmarks, labels tied to inputs, visible focus rings, and `prefers-reduced-motion` respected on every animation
- Every image with dimensions set, served as WebP with a fallback, and lazy below the fold

ANALYTICS, IF WANTED
- First-party: one small endpoint recording path, referrer host, a coarse device bucket and a salted visitor hash. No cookie, no third party
- Or nothing at all, which is a legitimate choice for a page like this

DEPLOYING
- Caddy with automatic TLS and the apex-to-www redirect decided one way and stated
- A systemd unit if there is a server component, or a plain static directory if there is not
- Cache headers: hashed assets immutable, the document itself short
- Security headers including a content security policy strict enough to actually forbid third-party scripts, which is easy here because there are none
- A one-line deploy script and instructions for pointing a domain at it

WHAT IT COSTS
- The honest comparison, in the README: this product is a few euros a month and gets a decent page live in twenty minutes. Building it takes an afternoon and then it is free, it is yours, it has no export problem, and nobody can change the terms. If the page is going to be edited by somebody who does not write code, keep the data file simple enough that they can

WHAT MATTERS MOST
Speed and the form. Get the page under 100KB and make the form work without JavaScript before adding a single animation. A one-page site that is slow or whose contact form silently fails has failed at the only two things it does.

Give me the repository, the data file with example content, the Caddy configuration, the deploy script, and a README with the domain setup written out step by step.