ci(website): deploy the site to Cloudflare Workers instead of GitHub Pages

.github/workflows/build.yml

@@ -1100,7 +1100,7 @@ jobs:
         # Only a stable release changes what /releases/latest resolves to, so a
         # pre-release would rebuild the site to byte-identical output.
         #
-        # Dispatched against main on purpose: the github-pages environment only
+        # Dispatched against main on purpose: the website environment only
         # permits `main` to deploy, so docs.yml's old `on: release` trigger ran
         # with a tag ref and failed its deploy every time. See docs.yml.
         if: ${{ steps.release.outputs.is_prerelease == 'false' }}

.github/workflows/docs.yml

@@ -18,8 +18,8 @@ on:
   #
   # That rebuild is a `workflow_dispatch` fired by build.yml once the release is
   # published, NOT an `on: release` trigger. A release event runs with
-  # github.ref = refs/tags/vX.Y.Z, and the github-pages environment only allows
-  # `main` to deploy, so the deploy job failed on every stable release (it never
+  # github.ref = refs/tags/vX.Y.Z, and the deploy environment only allows `main`
+  # to deploy, so the deploy job failed on every stable release (it never
   # surfaced earlier because pre-releases skipped the build entirely). Dispatching
   # against main both satisfies that policy and publishes main's docs rather than
   # the release branch's older snapshot.
@@ -119,33 +119,41 @@ jobs:
             exit 1
           fi
       - name: Upload artifact
-        uses: actions/upload-pages-artifact@fc324d3547104276b827a68afc52ff2a11cc49c9 # v5.0.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
+          name: website-build
           path: website/build
-          # v4 stopped bundling dotfiles by default. Nothing in website/build
-          # is hidden today — the deployed artifact from 7c4827b9 was checked
-          # and contains no dot-entries at all, so this changes nothing now
-          # (Docusaurus only writes a `.nojekyll` on its own `deploy` command,
-          # not in a plain `build`). It is set so that stays true by accident
-          # rather than by omission: a `.well-known/` for domain verification
-          # or security.txt is the realistic case, and under the v4+ default it
-          # would vanish from the deploy silently. `.git` and `.github` are
-          # excluded regardless of this flag.
+          retention-days: 7
+          # Off by default since v4. Nothing in website/build is hidden today,
+          # but a `.well-known/` for domain verification or security.txt is the
+          # realistic case, and it would vanish from the deploy silently.
           include-hidden-files: true
 
+  # The site is a Cloudflare Worker with static assets only (website/wrangler.jsonc).
+  # The token lives in the `website` environment, which only `main` may deploy
+  # from, so neither a PR nor a dispatch on another branch can read it.
   deploy:
-    name: Deploy to GitHub Pages
+    name: Deploy to Cloudflare
     runs-on: ubuntu-latest
     needs: build
     if: >-
       (github.event_name == 'push' && github.ref == 'refs/heads/main')
       || (github.event_name == 'workflow_dispatch' && github.ref == 'refs/heads/main')
     environment:
-      name: github-pages
-      url: ${{ steps.deployment.outputs.page_url }}
-    permissions:
-      pages: write
-      id-token: write
+      name: website
+      url: https://getopenscreen.com
     steps:
-      - id: deployment
-        uses: actions/deploy-pages@cd2ce8fcbc39b97be8ca5fce6e763baed58fa128 # v5.0.0
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+          sparse-checkout: website/wrangler.jsonc
+          sparse-checkout-cone-mode: false
+      - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: website-build
+          path: website/build
+      - name: Deploy
+        working-directory: website
+        env:
+          CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_API_TOKEN }}
+        run: npx --yes wrangler@4.145.0 deploy

technical-documentation/engineering/ci-workflows.md

@@ -13,7 +13,7 @@ flowchart TD
         CI --> BuildSmoke[Vite build]
         CI --> Semantic[semantic PR title]
         Docs["docs.yml<br/>website changes or dispatch"] --> DocsBuild[website build artifact]
-        DocsBuild --> DocsDeploy[GitHub Pages deploy on main]
+        DocsBuild --> DocsDeploy[Cloudflare deploy on main]
     end
 
     subgraph T2["Tier 2 — native and release builds"]
@@ -138,4 +138,4 @@ Each workflow needs variables or credentials, and where it checks for them decid
 
 Release platform artifacts have 30-day Actions retention. `publish-release` copies the NSIS installer, two architecture-specific DMGs when present, and Linux packages into the GitHub release. It intentionally does not download the independently uploaded Store artifact. A stable published release then fans out to Homebrew, WinGet, and AUR.
 
-Diagnostic artifacts remain Actions-only for 14 days. Whisper archives remain Actions-only for 30 days and are used as a binary-refresh output rather than being downloaded by the release workflow. The website build artifact flows only into GitHub Pages deployment.
+Diagnostic artifacts remain Actions-only for 14 days. Whisper archives remain Actions-only for 30 days and are used as a binary-refresh output rather than being downloaded by the release workflow. The website build artifact flows only into the Cloudflare deployment, which reads its token from the `website` environment, restricted to `main`.

website/.gitignore

@@ -5,6 +5,8 @@ node_modules/
 build/
 .docusaurus/
 .cache-loader/
+# wrangler dev's local state
+.wrangler/
 
 # Generated files
 .bundle-stats

website/README.md

@@ -1,13 +1,13 @@
 # OpenScreen docs site
 
-Docusaurus 3 site. Lives in `website/` inside the monorepo. Deployed to GitHub Pages via `.github/workflows/docs.yml`.
+Docusaurus 3 site. Lives in `website/` inside the monorepo. Deployed to Cloudflare (a Worker serving static assets, `wrangler.jsonc`) via `.github/workflows/docs.yml`. `www-redirect/` is a separate Worker that sends www.getopenscreen.com to the apex; it is deployed by hand.
 
 ## Develop
 
 ```sh
 cd website
 npm install
-npm run dev      # http://localhost:3000/openscreen/
+npm run dev      # http://localhost:3000/
 ```
 
 ## Build
@@ -25,6 +25,4 @@ npm run typecheck
 
 ## Notes
 
-- Site URL: <https://getopenscreen.github.io/openscreen/>
-- The base URL is `/openscreen/` because the org-level Pages site hosts multiple projects.
-- Docs migration from `docs/` (root) → `website/docs/` happens in a follow-up PR. For now only the intro page is published.
+- Site URL: <https://getopenscreen.com/>, served from the domain root (`baseUrl: "/"`).

website/docusaurus.config.ts

@@ -333,13 +333,13 @@ export default async function createConfig(): Promise<Config> {
 		tagline: "A free, open-source screen recorder and editor.",
 		favicon: "img/logo-icon.png",
 
-		// Pages serves this from the custom domain's root, not from
-		// getopenscreen.github.io/openscreen/, so baseUrl has to be "/" — a project
-		// baseUrl would prefix every asset URL with a path the server has nothing at.
+		// Served from the custom domain's root, so baseUrl has to be "/": a
+		// project baseUrl would prefix every asset URL with a path the server has
+		// nothing at.
 		url: SITE_URL,
 		baseUrl: "/",
 
-		// Every page is emitted as <route>/index.html, and GitHub Pages 301s the
+		// Every page is emitted as <route>/index.html, and Cloudflare redirects the
 		// extensionless form to the trailing-slash one. Leaving this unset makes
 		// Docusaurus advertise the pre-redirect URL in both <link rel="canonical">
 		// and sitemap.xml, so every indexed URL costs a crawler an extra hop to a
@@ -439,11 +439,13 @@ export default async function createConfig(): Promise<Config> {
 					// for a renamed tag: a translated build has no blog, so that copy
 					// would be the only thing under /fr/blog/, a page nothing links to,
 					// redirecting a French URL to an English one. /fr/discord/ would be
-					// a second, unlinked copy of the invite redirect, and crawlers read
-					// llms.txt and robots.txt at the root only. The rest (img, video)
-					// stays: pages resolve images through the locale's baseUrl.
+					// a second, unlinked copy of the invite redirect, crawlers read
+					// llms.txt and robots.txt at the root only, and Cloudflare reads
+					// _headers there only, serving any other copy as a public file. The
+					// rest (img, video) stays: pages resolve images through the locale's
+					// baseUrl.
 					if (LOCALE !== "en") {
-						for (const entry of [BLOG_PATH, "discord", "llms.txt", "robots.txt"]) {
+						for (const entry of [BLOG_PATH, "discord", "llms.txt", "robots.txt", "_headers"]) {
 							await rm(path.join(outDir, entry), { recursive: true, force: true });
 						}
 					}

website/i18n/TRANSLATING.md

@@ -6,7 +6,7 @@ The site is translated into `fr`, `es`, `pt-BR`, `ja`, `zh-CN`, `zh-TW` and `de`
 
 ## What is translated, and what is not
 
-- **Translated:** everything but the blog: the landing page, `/download/`, the docs, the comparison, platform and feature pages (`src/pages/**/*.mdx`), the navbar, the docs sidebar, the footer and the theme (404 page, pagination, "last updated"…). The translated 404 page only shows after a click inside the site: GitHub Pages answers every missing URL, `/fr/…` included, with the English one.
+- **Translated:** everything but the blog: the landing page, `/download/`, the docs, the comparison, platform and feature pages (`src/pages/**/*.mdx`), the navbar, the docs sidebar, the footer and the theme (404 page, pagination, "last updated"…). A missing URL under `/<locale>/` gets that locale's 404 page.
 - **English only:** the blog, a dated development journal. A translated build does not contain it, and links to it from translated pages go to the English blog (see below).
 - **The build enforces it:** a locale missing any doc or `.mdx` page fails `npm run build`, because Docusaurus would otherwise publish the English file under `/<locale>/` marked as your language.
 

website/static/_headers

@@ -0,0 +1,7 @@
+# Read by Cloudflare at the root of the deployed build (website/wrangler.jsonc).
+# Docusaurus puts a content hash in the name of every file under assets/, so a
+# cached copy can never go stale.
+/assets/*
+  Cache-Control: public, max-age=31536000, immutable
+/:locale/assets/*
+  Cache-Control: public, max-age=31536000, immutable

website/wrangler.jsonc

@@ -0,0 +1,21 @@
+// Cloudflare Workers config for getopenscreen.com, deployed by .github/workflows/docs.yml.
+// Assets only, no Worker script: Cloudflare serves static assets without counting
+// them against the Workers request quota.
+{
+	"name": "getopenscreen-website",
+	// An identifier, not a credential: the API token in CI is the secret.
+	"account_id": "49f0ef8f15d646deb80a7e7d675a4910",
+	"compatibility_date": "2026-10-01",
+	"assets": {
+		"directory": "./build",
+		// The nearest 404.html, so /fr/missing/ gets the French page, not the English one.
+		"not_found_handling": "404-page"
+	},
+	// A route on the proxied apex, not a custom domain: Cloudflare will not attach a
+	// custom domain over existing DNS records (error 100117). With a route, the move off
+	// GitHub Pages was only switching those records to proxied, with no gap.
+	"routes": [{ "pattern": "getopenscreen.com/*", "zone_name": "getopenscreen.com" }],
+	// The site has one address; a workers.dev copy would be a duplicate of it.
+	"workers_dev": false,
+	"preview_urls": false
+}

website/www-redirect/index.js

@@ -0,0 +1,11 @@
+// www.getopenscreen.com only: every request goes to the same path on the apex, as
+// GitHub Pages did. Deployed by hand (`npx wrangler deploy` in this folder); it has no
+// reason to change with the site.
+export default {
+	fetch(request) {
+		const url = new URL(request.url);
+		url.protocol = "https:";
+		url.hostname = "getopenscreen.com";
+		return Response.redirect(url.toString(), 301);
+	},
+};

website/www-redirect/wrangler.jsonc

@@ -0,0 +1,11 @@
+// Redirects www to the apex (index.js). The site itself is ../wrangler.jsonc.
+{
+	"name": "getopenscreen-www-redirect",
+	"account_id": "49f0ef8f15d646deb80a7e7d675a4910",
+	"main": "index.js",
+	"compatibility_date": "2026-10-01",
+	"workers_dev": false,
+	"preview_urls": false,
+	// A route, like the site's: the www record only has to be proxied.
+	"routes": [{ "pattern": "www.getopenscreen.com/*", "zone_name": "getopenscreen.com" }]
+}