# Three product sites from one Astro codebase, priced at build time

> How bweze.com, sms.bweze.com and domains.bweze.com build from one repo, read every price from the live price list, and stay readable with JavaScript off.

Source: https://bweze.com/blog/three-sites-one-codebase/ · Updated 2026-11-17

BWEZE sells three things that people search for separately: hosting, SMS and domains. So there are three sites, `bweze.com`, `sms.bweze.com` and `domains.bweze.com`, each with its own home page, pricing, guides and blog. They share a header, a footer, a design and most of their code, so they come from one Astro project.

## One codebase, three builds

An environment variable, `BWEZE_SITE`, picks the site at build time: `deploy`, `sms` or `domains`. Everything that differs is data keyed by that value: the site's name and origin, its search description, the meters it is billed by, its topics. Content carries a `site` field too, so an SMS guide only builds on the SMS site. Each build writes to its own folder and is uploaded to its own project.

The sites are static HTML on Cloudflare Pages. BWEZE uses Cloudflare's network in front of its own services as well, and we say so rather than pretend otherwise.

## A price is never typed

Every price on these sites comes from BWEZE's public price list, `GET /v1/pricing`, read once when a site is built. The pricing page, the rate tables, the "typical month" example and the cost tables in posts like [what a production app costs](/blog/what-a-production-app-costs/) are all computed from it.

A deploy build sets `SITES_REQUIRE_LIVE_PRICING=1`, and if the API cannot be reached the build fails instead of publishing old numbers. A local build falls back to a seed copy of the list in the repository, and the pricing page says that it did.

Articles follow the same rule. A sentence that mentions a rate names a meter, and the build fills in the number. In this post, app memory is 20 RWF per GiB-hour because the list says so today; if the list changes, the next build changes the sentence. Comparison tables use tokens in their frontmatter for the same reason.

## Pages that read without JavaScript

Every page is complete HTML. The menus are `<details>` elements, so they open without a script. The animations, including the isometric drawing in the bweze.com hero, are SVG drawn in code and moved by CSS classes; JavaScript only adds and removes the classes. With JavaScript off or reduced motion on, each drawing rests in a finished state.

Fonts are self-hosted, so a page makes no requests to anyone else. Analytics runs only after a visitor allows it in the cookie banner.

## Written for search engines and assistants

Each page has one title, a description and a canonical address, plus structured data for what it is: `Organization` and `WebSite` everywhere, `Service` with offers in RWF on product pages, `FAQPage` where there are questions, `BlogPosting` on posts and `DefinedTerm` in the glossary.

Each site also publishes `llms.txt`, an index for language models, and `llms-full.txt`, every guide, comparison and post in one file. Every article has a Markdown twin at its address plus `.md` ([this post's](/blog/three-sites-one-codebase.md)), with prices already filled in. An assistant that reads a page gets the same facts a person does.

## Facts carry their sources

The content schema refuses a guide without sources, and a comparison needs at least two, plus the date the competitor's facts were checked. That makes the rule from the plan for these sites, that each fact is checked against a source before it ships, something the build enforces instead of something we remember.

## Sources

- [Astro docs: Content collections](https://docs.astro.build/en/guides/content-collections/)
- [llms.txt proposal](https://llmstxt.org/)
- [BWEZE price list (GET /v1/pricing)](https://api.bweze.com/v1/pricing)
- [Cloudflare Pages docs](https://developers.cloudflare.com/pages/)
