Website Change Monitor — Hash & Text Diff

Watch a list of URLs for content changes — SHA-256 hash, optional CSS selector, and a difflib text diff. No browser. No paid LLM. Paste URLs, or chain from Sitemap URL Discovery / URL Status Checker datasets. Snapshots persist in a named (or ID) key-value store so scheduled runs can compare against the last check. Default memory: 256 MB. Failed network checks are free by default.

What you get

• 🔐 Content hash — SHA-256 of normalized page text (script/style stripped)
• 🧩 Optional CSS selector — hash only main / article / #content
• 📝 Text diff — stdlib difflib unified diff + lines added/removed (no LLM “explanation”)
• 🧺 Persistent snapshots — named KV store (or explicit store ID) across runs
• 🔌 Chain-friendly — same { "url": ... } shape as Sitemap / URL-status outputs
• 💸 PPE — low start + per successful check + extra only when hash changes; failures free by default
• 💾 Light — HTTP GET only, target ≤512 MB

Measured results

Local + private cloud benches (2026-10-02 Asia/Taipei). PPE locked after cloud margin benches — see docs/PRICING.md.

Test	Result
Local smoke (2 URLs, first / second / forced-change)	first×2 → unchanged×2 → changed×1 + page-changed charge; peak 78 MB
Local errors (1 ok + 1 bad DNS)	charged 1 check only (fail free); peak 78 MB
Cloud smoke rK92EHmHttVUrjPhY / dncJ8mOaNHB5OgzUh (2 URLs, 256 MB, build 0.1.2)	SUCCEEDED; peak 86 MB; settled ~$0.00037
Cloud fail-free ZOyhRn1x65VdFG73d	check×1 charged; settled $0.000120
Cloud bench P89Kp73Zbw8wYUv6e (5 URLs)	peak 88 MB; settled $0.000492 (~$0.000098 / URL)
Cloud changed h7LQ6fUdvbnNWOTl0	check×2 + changed×1; peak 84 MB; settled $0.000341

Use cases

• Scheduled content watch on a sitemap-derived URL list
• Competitor / pricing page change alerts (hash + diff, not marketing AI)
• Re-crawl trigger — only re-run doc/enrichment Actors when changed=true
• Post-hygiene monitor after URL Status Checker

How to use

1. Add URLs, and/or a Source dataset ID from Sitemap / URL-status.
2. Set a Snapshot store name (or ID) you will reuse on every schedule.
3. Optional: CSS selector, ignore regexes, output filter changed.
4. Click Start. Dataset + SUMMARY / OUTPUT in the run KV; snapshots in the snapshot store.

Input example

{
  "urls": [
    { "url": "https://example.com/" },
    { "url": "https://example.org/" }
  ],
  "snapshotStoreName": "website-change-snapshots",
  "storePreviousText": true,
  "maxUrls": 100,
  "outputFilter": "all"
}

Chain from Sitemap / URL-status:

{
  "datasetId": "<upstream-run-default-dataset-id>",
  "snapshotStoreName": "my-site-watch",
  "cssSelector": "main",
  "outputFilter": "changed"
}

Output example (one dataset item)

{
  "url": "https://example.com/",
  "finalUrl": "https://example.com/",
  "httpStatus": 200,
  "ok": true,
  "contentHash": "b94d27b9…",
  "previousHash": "a3f1…",
  "changed": true,
  "isFirstSeen": false,
  "diffSummary": {
    "isFirstSeen": false,
    "linesAdded": 2,
    "linesRemoved": 1,
    "unifiedDiff": "--- previous\n+++ current\n@@ …",
    "truncated": false
  },
  "checkedAt": "2026-10-02T01:00:00+00:00",
  "errorClass": null
}

Key-value store records (run)

Key	Content
SUMMARY	Counts: totalChecked, changed, firstSeen, failed, charged, peakMemoryMb
OUTPUT	Same summary

Snapshots are written to the snapshot store (not the run default KV), keyed as snap:<sha256(url)>.

Pricing

Pay per event (private lock; see docs/PRICING.md):

Event	Price
Actor start (Apify synthetic)	$0.001 / GB (1 GB min)
Page checked (primary)	$0.0015 per successful fetch+hash
Page changed	$0.008 when hash differs from previous snapshot

Failed checks (DNS / timeout / network) are not charged unless you enable chargeFailedChecks. First-seen pages charge page-checked only (not page-changed).

Chaining

Sitemap URL Discovery → URL Status Checker → Website Change Monitor
                                              └─ on changed → re-run doc/enrichment

Limits

• HTTP only — no headless browser / JS rendering (keeps memory and CU low).
• Dynamic pages that inject clocks/CSRF into the main text may need ignorePatterns or a tighter cssSelector.
• Very large HTML is truncated at maxBodyBytes (default 2 MB).

License

Actor source: AGPL-3.0. Dependencies: httpx (BSD-3), selectolax (MIT), apify (Apache-2.0). See docs/LICENSE_REVIEW.md.