Baitless News

Proof of Concept URL : https://baitless-news.hypothetic.dev/

An experiment demonstrating Cloudflare Pages plus a Cloudflare Worker that rewrites daily headlines into concise, non-clickbait summaries with GPT-5-nano.

Repository layout

• web/public/ – Static frontend served via Cloudflare Pages. Displays summaries fetched from /api/news.
• worker/ – TypeScript Cloudflare Worker that pulls headlines and asks GPT-5-nano for calmer summaries.
• wrangler.toml – Worker configuration used by Wrangler for local development and deployment.

Frontend (Cloudflare Pages)

The web/public directory is deployable as-is. For local iteration you can run a static file server, or use Wrangler's Pages tooling once configured:

wrangler pages dev web/public

When you connect the repository to Cloudflare Pages, set the build command to null (no build) and output directory to web/public.

Worker API (/api/news)

The Worker now accepts GET requests and returns JSON shaped like:

{
  "stories": [
    {
      "id": "story-1",
      "title": "Calmer headline",
      "summary": "Non-clickbait summary text.",
      "sourceUrl": "https://..."
    }
  ]
}

It fetches the latest UK headlines from the Daily Express RSS feed, spoofing a realistic desktop Chrome user agent (and related headers) so the site treats requests like a normal browser. The worker parses both the RSS and article HTML with linkedom's DOMParser so it can iterate over Express' repeated div[data-mrf-recirculation="Link Content Paragraph"] blocks and recover proper article text before prompting GPT-5-nano for calmer rewrites. Network calls are made on demand; consider caching before shipping to production.

Required environment values

Store this in Cloudflare (or a local .dev.vars file when using wrangler dev):

• OPENAI_API_KEY – GPT-5-nano access token. If omitted the Worker will return raw RSS content without rewrites.

Local development

cd worker
npm install
# Optional: add OPENAI_API_KEY to .dev.vars before starting dev
npm run dev

The dev server binds to http://localhost:8787; hitting http://localhost:8787/api/news fetches the Express feed and, when configured, invokes GPT for rewrites. Running npm install pulls in the runtime dependency on linkedom, which supplies the DOMParser used to extract article bodies.

Deployment

1. Configure wrangler.toml with your preferred project name and add a route, e.g.

   name = "baitless-news-worker"
   main = "worker/src/index.ts"
   compatibility_date = "2024-04-02"
   
   routes = [
     { pattern = "baitless-news.hypothetic.dev/api/news", zone_name = "hypothetic.dev" }
   ]

2. Store the OpenAI secret in Cloudflare:

   wrangler secret put OPENAI_API_KEY

3. Deploy (Wrangler runs npm install --prefix worker automatically before bundling, so the Worker always ships with the linkedom dependency):

   npm run deploy

Next steps

• Add caching or scheduled jobs so GPT calls are reused instead of running on each request.
• Connect the Pages project and Worker route inside the Cloudflare dashboard.
• Add integration tests or monitoring before the live demo.