Watch Any Web Page for Changes With changedetection.io

There’s a specific flavor of guilt I know well: the tab I keep open to a project’s releases page, meaning to check whether the security fix shipped yet. Or the vendor advisory page I should be watching. Or the price of the thing I want, which I refresh at random and always seem to miss the drop on. I keep meaning to check these, and I never do — because “remember to manually check a web page on a schedule” is a job humans are catastrophically bad at.

So I stopped doing it. changedetection.io is a self-hosted watcher: you give it a page, it polls it on a schedule, and it pings your phone only when something actually changes. This is the fourth stop in the Self-Host the Apps You Actually Use series, and it’s the one that turned a nagging chore into something I genuinely never think about anymore.

Here’s the loop it runs so you don’t have to:

What changedetection.io is

changedetection.io is, at heart, a very disciplined robot that reloads web pages and compares them to last time. You add a watch — a URL plus how often to check it — and it stores a snapshot. On each check it fetches the page again, diffs it against the stored version, and if the parts you care about changed, it fires a notification. It’s open source, it’s light, and it’s genuinely one of those tools that pays for its keep the first week.

The classic uses are exactly the chores you keep forgetting: software release pages, security advisories, product prices, restock alerts, and terms-of-service or policy pages. Anything that lives at a stable URL and updates on its own clock is a candidate.

Deploy it

changedetection.io offers both a Docker install and a native pip install (it needs Python 3.10+). I run it natively in a small Proxmox LXC under systemd, but the Docker route is a single command if you’d rather.

# Native (Python 3.10+), pinned to a known release:
pip install changedetection.io==0.55.8
changedetection.io -d /var/lib/changedetection/data -p 5000

# Or Docker:
docker run -d --restart always -p 127.0.0.1:5000:5000 -v datastore-volume:/datastore --name changedetection dgtlmoon/changedetection.io

The real skill: pick endpoints a browser doesn’t need

Here’s the trick that keeps this running happily on a tiny container. changedetection.io can drive a full headless browser to render JavaScript-heavy pages — but that’s heavy, and you rarely need it. If the information you want is exposed as a JSON API, an RSS/Atom feed, or plain static HTML, you can fetch and parse it directly with no browser at all. It supports XPath, JSONPath, jq, and CSS filters under the Filters & Triggers tab, so you can pinpoint exactly the value that matters and ignore the churn around it.

That framing changes how you choose what URL to watch. Instead of watching a vendor’s pretty download page (JavaScript, ads, a rotating banner — all noise), watch the JSON or feed behind it. Here are the kinds of targets I reach for:

Every one of those is parseable without a browser, which is why five real watches sit comfortably on a 1 GB box.

The RSS trap that alerts you forever

The first time I watched an RSS feed unfiltered, my phone buzzed on every single check even though nothing new had been posted. The culprit: most feeds include a lastBuildDate (or similar) timestamp that’s regenerated every time the feed is served. To changedetection.io, that timestamp changing is a change. The fix is to filter down to just the part you care about — for a feed, an XPath selecting the item titles:

Push it to your phone with ntfy

Notifications run through Apprise, a library that speaks to dozens of services. My favorite for this is ntfy — dead simple push notifications you subscribe to by topic in a phone app. In a watch’s Notifications list, add an Apprise ntfy URL:

# public ntfy.sh, dedicated topic:
ntfy://homelab-releases

# or a self-hosted ntfy server:
ntfys://ntfy.homelab.lan/homelab-releases

To be sure it works end to end, I add a throwaway watch on a page that changes on every request, force a recheck, and confirm the ping lands on my phone. Then I delete the test watch. Proving the whole chain — detect → Apprise → ntfy → phone — beats assuming any one link works.

Watch the watcher

There’s a delicious irony here: a tool whose entire job is to notice things can itself die quietly and stop noticing anything, and you’d never know. So I point a black-box uptime probe at its web UI and alert if it stops answering — a dead-man switch, so a dead watcher pages me instead of failing silent. The companion playbook has the full deploy, five example watches, and the probe.

What’s next

That’s four apps that make the homelab genuinely more useful — private search, self-tagging bookmarks, an OCR’d document archive, and a tireless page-watcher. To close the series, we go one layer down and protect the thing all of it runs on: SMART disk-health trending, so a dying drive warns you weeks ahead instead of taking your data with it.

Related posts:

• Paperless-ngx: OCR Every Document You Own, Searchable — the previous post: make your documents findable.
• SMART Disk-Health Trending for Your Whole Fleet — the series finale: watch the disks under everything.
• Give Your Local AI Private Web Search With SearXNG — where the series started: your own private search engine.
• Uptime Kuma: Simple Homelab Monitoring — the friendly status-page companion to a change watcher.
• What Is Prometheus? — the metrics stack behind the dead-man probe.
• Tailscale Subnet Router for Your Homelab — reach an unauthenticated tool safely from anywhere.
• Create Your First Proxmox LXC Container — where the watcher lives in this build.