How SOSEI Rebuilds a Website in Minutes

Most people’s mental model of “AI-generated website” is a single ChatGPT prompt followed by a wall of generic markup. That’s not what happens here. A SOSEI rebuild is one agent that plans, designs and writes the site, with legal pages running alongside and code checking the result. The whole thing takes minutes for a typical small-business site — and the engineering choices that make it reliable are worth unpacking.

Stage 0: Scrape

Before any AI runs, the source site has to be understood. A parallel crawl — Cheerio over fetch, up to 50 pages, fewer when the site only answers a browser — reads the homepage links and the sitemap and extracts:

• Visible text, structured by heading hierarchy.
• Image references, with their natural dimensions.
• Brand: colour palette, fonts in use, logo SVG, favicon.
• Tracking codes: GA, GTM, Meta Pixel, Hotjar, LinkedIn Insight, TikTok Pixel, Plausible, Umami, Clarity.
• E-commerce detection: Shopify, WooCommerce, BigCommerce, Magento, Wix, Squarespace.

Every image — including the logo and favicon — is then re-fetched and uploaded to Supabase Storage. This is the step that makes downstream generation safe: no AI prompt ever references an external image URL that might fail at render time, and the rebuild keeps working if the original site goes offline.

Stage 1: Plan

The plan call receives the entire scrape and produces a structured site plan:

• The page list (Home, About, Services, Contact, plus industry-specific pages like “Portfolio” or “Menu”).
• The detected industry ID from a 163-entry catalog. This drives downstream design and copy choices — a dental clinic and a metal-fabrication shop get different colour palettes, typography registers, and section orderings.
• The recommended font pairing from a self-hosted pool of more than 200 faces, chosen to be the closest mood-match for the source site.
• A brand-style summary (formal / playful / authoritative / minimal / etc.).

Which model runs that call is an admin setting, and there is no quality tier to pick.

Stage 2: Voice card

Before any page is written, a separate, short AI call produces a frozen voice contract — the tone, rhythm, person, vocabulary, and recurring phrases of the source site. Pages are then written against that contract, including any a cheaper model types.

Why a separate step? Because models drift. Pages written without a shared voice contract end up sounding like different writers. Lock the voice once, before the pages are written, and the site reads like it came from the same author. See Preserving Your Brand During a Rebuild for the full picture.

Stage 3: Design system

Two AI calls build the visual layer. The first produces design tokens (colour, typography, radius, spacing, shadow), a mood descriptor, a style profile (“editorial”, “swiss”, “playful”), and a motion profile (“subtle”, “none”, “expressive”). The second composes the global header and footer HTML.

The tokens then run through a deterministic normalizeTokens pass that enforces WCAG contrast ratios (AAA 7:1 on body, AA 4.5:1 on muted) and fills any omitted tokens from the catalog palette. The AI cannot output an accessibility-violating design system because the renderer rejects it.

Stage 4: The pages

One agent writes the pages, one at a time. It writes the HTML and the CSS for each page, then looks at what it wrote. A cheaper model can do the mechanical typing. The agent is not handed a menu of JSON sections to fill in.

Code does not trust that markup blindly. It sanitizes the CSS, enforces contrast, and rejects a page that omits the markers it has to carry. A misclosed tag or a banned rule does not ship just because the model emitted it.

Stage 5: Legal pages (in parallel)

While the pages are being written, the six legal pages are written alongside, one call each: privacy policy, terms of service, cookie policy, impressum, accessibility statement, and the 404 page. They don’t depend on the per-page copy — only on the source-site metadata (company name, address, email) and language.

Each page comes back as HTML, not as one JSON document. A page that fails its checks is asked again. If one is still missing, the site can publish and the gap is flagged: an automatic retry fills it in, and the dashboard has a button for the same repair.

Stage 6: Host

Everything assembles into a hosted site: index.html, every inner page, style.css, _header.html, _footer.html, sitemap.xml, robots.txt (with explicit allow rules for all 18 major AI crawlers), llms.txt, and the legal pages. All saved into a versioned row in Postgres and served live at https://…/s/[slug]/….

How long a rebuild takes

There is no Fast or Best switch. A rebuild takes minutes, and there are not two products to pick between. Which model runs each stage is an admin setting, not a control on the dashboard.

The logo still gets a small vision pass — enough to pull two or three dominant brand colors from the image. A larger model would only add cost for that job.

What code checks, instead of a section menu

The old path compiled a JSON list of sections into HTML. A normal build does not do that. The agent writes the page, and code checks the result:

• Contrast. Body text is held to AAA 7:1, and muted text and accent-on-surface to AA 4.5:1. A palette that misses is corrected.
• CSS. The stylesheet is sanitized: no imports, no external images pulled in through CSS, and no fixed layer that covers the page with no way out.
• Markers. A page that omits the required component markers is rejected, rather than shipped and hoped over.

Why this matters for what you see on launch day

The result of all this engineering is the only outcome the customer actually cares about: a site that looks like itself, works on every device, scores high on every audit dimension, and ships in minutes instead of three months. Every step in the pipeline exists because removing it produced a measurable regression in some real-world rebuild.

If you’ve only ever seen “AI website builders” as a curiosity, the SOSEI rebuild is worth trying as a benchmark. Start a project with one URL and judge the output.