Static HTML on the edge
Astro renders every page to plain HTML at build time; Cloudflare serves it from Workers Static Assets. The reader gets HTML and a few kilobytes of CSS — almost no JavaScript runs.
A blog with an engine behind it
spintax.site is the support blog of spintax.net. It writes about producing content with AI and spintax at scale, and every number in it is measured on the open engine or checked against the live source, with the date. The site itself is small: static pages, search in the browser, cards drawn at build time, a surface built for AI readers, and a Russian twin on a host that Russia can reach.
The writing has one job: start from a real problem of content at scale — pages Google deindexes, mail that spam filters throttle, a catalog of a thousand near-identical descriptions, a model that writes the same sentence fifty ways — and measure rather than argue. Where the manual path runs out, that is where spintax.net picks up: the open engines, the n8n node, Spintax Studio, the MCP server.
The stack
A fork of 301.sh, the same kind of blog for a sibling product. A handful of pieces, none of them a server.
Content is Markdown in git
No CMS, no database, no login screen. An article is a file, an edit is a commit, a merge is the deploy — and every figure carries the date it was last checked.
Measured on the open engine
Variant counts, uniqueness, lint findings: whatever an article claims about spintax was run through @spintax/core, the MIT engine behind spintax.net, with the command in the text.
Search runs in your browser
Pagefind builds a static index at deploy and searches it client-side with WebAssembly. There is no search server and no per-query cost — the index is just more static files.
Share cards drawn at build
Every article image — and every social card — is generated from its title and tag: SVG, text traced to outlines, rendered to PNG, one layout per platform. Deterministic and reproducible.
A surface for AI readers
An llms.txt index, a Markdown twin of every page, content negotiation on the Accept header and an MCP server with the same two tools: an agent gets clean source instead of page chrome.
A Russian twin off Cloudflare
Cloudflare is blocked in Russia, so the Russian articles are served from ru.spintax.site on GitHub Pages, built from the same sources. The apex answers /ru/ with one redirect.
Where the articles go
When an article ships it lands in a small JSON feed, and the poster Worker of spintax.net takes it from there: one article per channel per tick, remembered in a KV namespace, never twice. The channels are the product’s own.
No analytics script, no tracking pixel. The feeds and the keyword map are public files anyone can read.
What it covers
Every article is filed under the problem it is really about. Pull a thread: