If you've ever written documentation for a living, you know the particular flavour of horror that comes from a vendor updating a page without saying a word. A deprecated flag. A changed default. A renamed permission. A URL slug that now 301s to something vaguely related but not actually the same. The kind of change that would be a one-line entry in a sensible changelog, if the changelog existed at all.
I've been quietly maintaining a collection of pages I need to watch for the better part of a decade - Google's Android Enterprise documentation, AMAPI references, OEM product pages, vendor announcement blogs, the occasional Wikipedia article that gets quietly edited when something interesting happens. For most of that time I've used various tricks and tools, and a fair amount of daily checks as part of my morning routines.
The trouble with this is me. Myself. Just a simple human reminding myself to read the docs on a regular basis.. and I miss things - either directly insomuch as forgetting to check in on a page I should frequent often, or indirectly by missing a change to wording within a paragraph. The latter being a subtle but critical function change when Google updates the default of a policy, for example.
So, in the spirit of Flash and MIKA, I built the thing I actually wanted.
DeltaWatch is a multi-tenant web change detection platform. You point it at a URL, it fetches the page, extracts the bits you care about, and tells you when something changes. That's the core of it. Everything else is about doing that well at scale, without noise, without fuss, and without having to babysit an installation in a NUC in your cupboard.
It's live right now at deltawatch.ing, the free tier has no time limit, and you can sign up in about fifteen seconds.
Existing tools occupy one of two camps. Commercial services like Visualping, Distill, and Wachete are polished but genuinely expensive at any usable scale - £10 to £15 a month for monitoring a few dozen pages with reasonable check frequency is the going rate, and their free tiers are either time-limited or limited to a handful of watches with hour-plus check intervals. The open-source alternative - changedetection.io - is fantastic, but it's a self-hosted service, which means it's my problem to keep running, update, back up, and debug at 2am when the Docker container decides it's had enough.
What I wanted was a hosted service that priced sensibly, had no time limit on the free tier, gave me the surgical content-extraction controls I need for the long-tail pages I actually care about, and didn't require me to become an SRE for my own monitoring stack.
So I built one.
The feature list is too long for an announcement post, so I'll stick to what I use personally every day.
Most monitoring tools let you pick a CSS selector and call it a job done. DeltaWatch lets you target the exact content you care about via CSS selectors, XPath, JSONPath, or regex - and combine them. Want to watch a specific <div> on a page, but only the changes inside the third <p>? Done. Want to watch a JSON API response but only specific keys? Done. Want to strip navigation, headers, footers, sidebars, and anything with display: none before the diff runs? One checkbox.
JSON responses are auto-detected and formatted with sorted keys, so changes in key ordering never trigger false alerts. PDFs up to 20 MB are text-extracted and diffed like any other content. There's a workspace-level "strip blank lines" toggle for the sites that love to insert random whitespace. All of this is there because I hit every single one of these edge cases within the first week of using the platform on my own pages.
Content extraction gets you surgical targeting. The conditions engine filters out the noise that slips through anyway. Eight operators (contains, does not contain, equals, greater than, regex match, and friends), AND/OR logic, and change-percentage thresholds so you only get alerted when the page changes by more than, say, 5%. Every condition composes with every other condition. It's the difference between getting one useful alert a week and getting twelve useless alerts a day.
Email is built in. Everything else is a URL away: Slack, Discord, Telegram, Microsoft Teams, and generic webhooks. Add channels per watch, per tag, or at the workspace level. Every channel is available on every plan, including the free tier, because pay-walling notifications on a monitoring product is absurd.
This solves a real and annoying issue for me - missing new content because instead of a page changing, a new page is added to the website.
Sites lets you model an entire domain as a connected graph or table. You point DeltaWatch at a root URL, it crawls the hostname, builds a graph of pages and the links between them, and gives you an interactive network or table view of the whole site. You can see what links to what, which pages have drifted recently, and which pages are orphaned or rarely updated.
From the graph (or a table view, if you prefer rows), you can promote any discovered page into a real watch with one click - or bulk-promote a dozen at a time. Promotion routes through the standard watch creation flow, so plan limits, duplicate detection, and interval enforcement all still apply. There's also an optional auto-watch toggle that turns every newly discovered page into a watch as the crawler finds it, which is excellent for small focused sites and absolutely terrifying for anything the size of a Google docs property.
Sites also does recrawl scheduling. When a watch detects a change, the parent site queues an automatic re-crawl (debounced at 12 hours site-wide to avoid storms), so the data stays current with the pages you care about. As a fallback, every site gets a scheduled full recrawl roughly every 7 days with ±40% per-site jitter, so abandoned sites still stay fresh and the fleet doesn't all crawl at once.
Sites is currently a Sentinel-tier feature - more on pricing below.
I had eight years-worth of links to bring across when I started using this in earnest, so import had to work from day one. DeltaWatch imports from six formats:
Every import shows a preview before committing, enforces plan limits, and runs as a background job so large imports don't time out.
Workspaces are per-team, per-project, or per-client containers. Each workspace has its own plan, its own billing, its own users, and its own notification defaults. I run a personal workspace for my own monitoring, a bayton.org workspace for the documentation side, and a handful of client workspaces that are billed separately. You only pay for the workspaces that need paid features.
Every workspace can generate a full ZIP backup in the background - watches, tags, settings, and snapshot history - and download it via a time-limited link. No lock-in: if you ever want to leave, the exit door is right there.
For those of you who like a peek behind the curtain, here's the architecture.
Frontend is React 19 with Vite and Tailwind CSS 4. The Watchlist, site graph, diff viewer, and admin panels are all SPA components. The site graph uses a force-directed canvas simulation with viewport culling, zoom-based label visibility, and adaptive force parameters so it stays responsive at 1,000+ nodes. No D3 in the main bundle; it's a lightweight hand-rolled force simulation that weighs less than most charting libraries.
Backend is Express 5 with a comprehensive REST API. It runs in two deployment modes: as a Netlify full-stack application (SPA + serverless functions) lighter instances, and as a standalone Node.js process with a strongly-consistent PostgreSQL backend. The standalone build supports a split deployment topology where HTTP serving and background work run as separate processes, isolating API latency from crawl and watch-check load.
Storage is a pluggable backend abstraction. There are four backends - Netlify Blobs, Directus, a generic KVP-over-JSONB database store, and a native-columns Postgres store that uses typed columns and expression indexes instead of JSONB blobs for the hot tables. The native backend is what runs on the VPS in production, and it's materially faster than the KVP equivalent for listing, filtering, and bulk operations.
Watch check pipeline is durable. Every check runs as a persisted job with claim tokens, stale-job recovery, and a 10-minute timeout watchdog. One watch per job means a slow target can't strand a batch of twenty; the scheduler just moves on and comes back to it on the next tick. Imports, exports, and workspace deletes follow the same job-queue model, which means they survive process restarts and serverless context freezes.
The site crawler uses BFS with robots.txt support, rate limiting, checkpointing, a configurable edge cap, and - crucially - the same SSRF and redirect validation path as regular watch checks. A crawl can't reach anywhere a watch couldn't.
Platform health monitoring runs every 5 minutes on the worker process and captures disk usage, Node memory, database pool pressure, watch-check failure rate, crawl job health, scheduler miss detection, and API p95 latency. Alerts only fire on state transitions (ok → warn → critical) with a configurable cooldown, and there's a recovery notification when everything returns to ok. It's the kind of operational visibility you'd expect from a commercial SaaS, because that's what this is.
I wanted the pricing to be obvious, cheap, and sustainable. Three tiers:
That's it. No seat pricing. No watches-per-month metering. No "contact us for Enterprise". Per-workspace billing, so you can keep personal monitoring free while running a paid workspace for client work side-by-side.
For context, the commercial competitors want £10–15/month for a similar watch count with 5-minute checks. DeltaWatch does it for £6, and the free tier is actually usable rather than a trial.
A few things it deliberately isn't:
deltawatch.ing is live. Sign up, add your first watch in about thirty seconds, and the diff viewer will show you your first change the moment the content shifts.
The docs are at deltawatch.ing/docs, the release notes live at deltawatch.ing/releases, and there's a live status page at deltawatch.ing/status that reads straight from the production health endpoint so you can see what the platform's doing right now.
If you spot a bug, a content extraction edge case, a site that breaks the crawler, or a feature you miss from another tool, I want to hear about it. There's a reply-by-email link below, or reach out directly.
Happy monitoring!