I built a $0/month AI news aggregator with Next.js, GitHub Actions and IndexNow

alanaicode — Sun, 09 Aug 2026 06:48:40 +0000

Every morning I used to open a dozen tabs — Hacker News, arXiv, a few vendor blogs, two newsletters — just to figure out what actually happened in AI the day before. It took 30 minutes and I still missed things.

So I built a pipeline that does it for me, and it costs $0/month. Here is the whole architecture.

The stack

Layer	Choice	Why
Collection	GitHub Actions (cron)	Free minutes on public repos, no server to babysit
Processing	Plain Python (stdlib only)	Zero dependency drift, cold start is instant
Storage	JSON committed to the repo	Git is the database and the changelog
Rendering	Next.js static export	Pre-rendered HTML, perfect Lighthouse scores
Hosting	Vercel free tier	Push to deploy, global CDN included
Indexing	IndexNow	New URLs hit Bing/Yandex in seconds, not days

The result is live at AI Nexus Daily if you want to see the output before reading the how.

1. Collection runs in CI, not on a server

The single biggest cost saving is refusing to run a server. A scheduled workflow is enough:

name: build
on:
  schedule:
    - cron: '0 10 * * *'
  push:
    branches: [master]

jobs:
  build:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: python generate.py
      - name: Commit results
        run: |
          git config user.name "bot"
          git config user.email "bot@users.noreply.github.com"
          git add -A
          git diff --staged --quiet || git commit -m "chore: daily rebuild"
          git push

Public repositories get unlimited Actions minutes. A daily job that runs for 40 seconds will never cost anything.

2. Commit the data, don't host a database

generate.py writes data/items.json and commits it. That gives you three things for free:

Versioning — git log data/items.json is a full audit trail of every change.
Rollback — a bad scrape is one git revert away.
Deploy trigger — the commit itself makes Vercel rebuild. No webhook, no token.

The moment you reach for Postgres for a few thousand rows of read-mostly JSON, you have signed up for a monthly bill and a backup strategy you don't need.

3. Static export, not SSR

Feed content changes once a day. Server-side rendering it on every request means paying (in latency and in function invocations) for work that is identical 40,000 times in a row.

// next.config.js
module.exports = {
  output: 'export',
  trailingSlash: true,
}

Static export also means the site keeps serving even if the build pipeline breaks — the last good HTML is still on the CDN.

4. The part most people skip: telling search engines

Getting crawled is the actual bottleneck for a new content site. Sitemaps are passive — you publish one and hope. IndexNow is active: you ping the endpoint and Bing, Yandex and Seznam come fetch the URL, usually within minutes.

Two requirements that cost me an afternoon:

The key file must be reachable at the domain root. On Next.js that means public/<key>.txt, not the repo root. Putting it in the repo root gives you a confident 404.
The key file content must be exactly the key, no trailing newline drama, no HTML wrapper.

import json, urllib.request

payload = {
    "host": "ainexusdaily.vercel.app",
    "key": "YOUR_KEY_HERE",
    "keyLocation": "https://ainexusdaily.vercel.app/YOUR_KEY_HERE.txt",
    "urlList": ["https://ainexusdaily.vercel.app/"],
}
req = urllib.request.Request(
    "https://api.indexnow.org/indexnow",
    data=json.dumps(payload).encode(),
    headers={"Content-Type": "application/json; charset=utf-8"},
)
print(urllib.request.urlopen(req).status)  # 200 or 202

Google does not participate in IndexNow, so keep the sitemap for them. But getting indexed on Bing the same day still beats waiting a week for everyone.

What I would do differently

Start with the static export. I began with SSR out of habit and spent a day removing it.
Commit the data from day one. My first version regenerated everything on each run, so I had no way to see what changed — which turned out to be the interesting part.
Add IndexNow before writing any content. Indexing latency compounds; a page published on day 1 and indexed on day 8 lost a week of traffic it will never get back.

Cost breakdown

Item	Monthly
GitHub Actions (public repo)	$0
Vercel Hobby	$0
Domain (`.vercel.app` subdomain)	$0
Database	none needed
Total	$0

The whole thing has been running unattended for weeks. If you want to see what it produces daily, it is at ainexusdaily.vercel.app.

Happy to answer questions about any layer of this in the comments — especially the IndexNow part, since the docs are thin and the failure mode is silent.

DEV Community: alanaicode