RSS & Atom Feed Monitor - New Items Only + AI Digest

Monitor any number of RSS 2.0, Atom and RSS 1.0 (RDF) feeds and get only the items you have not seen before. Put it on an Apify Schedule (hourly, daily...) and every run delivers just the fresh entries - ready for Slack, email, Google Sheets, Make, n8n, Zapier or your AI agent via MCP.

Optional: let Claude write a 1-2 sentence summary per item and/or one Markdown digest per run, using your own Anthropic API key.

Why this feed monitor

• New items only, across runs. Seen item IDs are remembered per feed in a key-value store in your account. No duplicates in your downstream tools.
• Polite and cheap. Uses HTTP conditional requests (ETag / If-Modified-Since). Unchanged feeds return 304 Not Modified and cost almost nothing.
• Handles real-world feeds. RSS 2.0, Atom, RSS 1.0/RDF, content:encoded, Dublin Core dates, enclosures (podcasts), HTML stripped to clean text. Correctly decodes Shift_JIS and EUC-JP feeds that break many parsers (common on Japanese sites), and timezone-less Japanese dates.
• Keyword filters (include / exclude, case-insensitive, works with Japanese and Chinese text) and an age filter.
• Budget-safe. Respects your "Maximum cost per run". If the limit is hit, undelivered items stay pending and arrive on the next run - nothing is silently lost.
• AI optional, bring your own key. Model usage is billed by Anthropic directly to you. Default model claude-haiku-4-5 (cheapest); you can choose any Claude model.

Use cases

• Competitor blogs, changelogs and release notes -> Slack channel
• Security advisories (vendor and government feeds) -> ticket queue
• News monitoring for a brand or topic with keyword filters -> daily AI digest by email
• Podcast feed watcher (enclosure URLs) -> transcription pipeline
• Japanese government / corporate press-release feeds -> English digest (set Summary language to English)

Input

Field	Description
feedUrls	List of feed URLs (required).
onlyNewItems	Remember delivered items and output only unseen ones (default true).
firstRunBehavior	emit-all (default) outputs current items on the first run; baseline-only just remembers them.
stateStoreName	Key-value store that holds the memory. Use different names for independent monitors.
maxItemsPerFeed	Max items per feed per run (default 50). Older items beyond this limit in the same feed snapshot are marked as seen.
publishedWithinDays	Skip items older than N days.
includeKeywords / excludeKeywords	Keyword filters on title, summary and categories.
includeContentHtml	Also output full HTML content when available.
aiMode	none, per-item, digest, or both.
anthropicApiKey	Your Anthropic API key (secret input, needed only for AI).
aiModel, aiLanguage, aiInstructions	Model ID, output language, extra instructions.

Example:

{
  "feedUrls": ["https://hnrss.org/frontpage", "https://github.blog/feed/"],
  "onlyNewItems": true,
  "includeKeywords": ["security", "pricing"],
  "aiMode": "digest",
  "anthropicApiKey": "sk-ant-...",
  "aiLanguage": "English"
}

Output

One dataset item per delivered feed entry:

{
  "feedUrl": "https://github.blog/feed/",
  "feedTitle": "The GitHub Blog",
  "feedFormat": "rss2",
  "itemId": "https://github.blog/?p=99381",
  "title": "Hack the World: Why hackathons are still the best place to learn to build",
  "link": "https://github.blog/developer-skills/career-growth/hack-the-world-...",
  "publishedAt": "2026-10-09T16:00:00.000Z",
  "author": "Ed Summers",
  "categories": ["Career growth", "Developer skills"],
  "summary": "The barriers to building software have collapsed. ...",
  "aiSummary": null,
  "enclosures": [],
  "isNew": true,
  "detectedAt": "2026-10-12T12:58:11.580Z"
}

The run's key-value store also contains:

• RUN_REPORT - per-feed status (ok, not-modified, error), item counts and detected encoding.
• DIGEST.md - the AI digest (when aiMode is digest or both).

Pricing (pay per event)

You pay only for results:

Event	Price
Feed item delivered	$0.0008 per item ($0.80 per 1,000)
AI item summary	$0.002 per item (+ your own Anthropic usage)
AI digest	$0.01 per digest (+ your own Anthropic usage)

A run where nothing changed (all feeds 304 Not Modified) delivers no items, so you pay only the small Apify start fee.

Scheduling

Create an Apify Schedule (e.g. every hour) with the same input. Because the memory is stored in your account, each run only returns what is new since the previous run. Connect an integration (Slack, email, webhook, Google Sheets) to the "run succeeded" event to push new items automatically.

Responsible use

This Actor only reads feeds you list. Feeds are published for automated reading, but some publishers limit how their content may be reused - make sure your use of the content complies with each publisher's terms. The Actor sends a descriptive User-Agent, uses conditional requests, and fetches each feed once per run with low concurrency. It does not bypass logins, paywalls or bot protection.

FAQ

Does it scrape article pages? No. It reads only the feed itself. If the feed contains full content, enable includeContentHtml.

Where is my Anthropic key stored? It is a secret input: Apify stores it encrypted and it is only used to call the Anthropic API during your run.

A feed returns an error. Check RUN_REPORT for the HTTP status. Other feeds in the same run are still processed.

Japanese feed shows garbled text in other tools. This Actor detects the charset from the HTTP header and the XML declaration (Shift_JIS, EUC-JP, UTF-8 and others) and always outputs UTF-8.

About

Developed and maintained by koban-tools, with AI-assisted development (Claude). The code is tested against real RSS, Atom and RDF feeds before each release. Questions and bug reports are welcome via the Issues tab (English or Japanese).