RSS to Summary Digest

Fetches one or more RSS / Atom feeds, extracts each item, and produces a concise summary per item as a structured dataset.

How it works

1. Fetches each feed URL (follows redirects, 30s default timeout, browser-like User-Agent).
2. Parses RSS 2.0, Atom, and RDF feeds with a tolerant XML parser.
3. For each item it produces a summary using one of two methods:
   • Extractive (default, free): a deterministic, dependency-free summarizer that selects the leading sentences of the item content. No API key and no paid LLM calls — the only network requests are the RSS/Atom feed fetches themselves.
   • LLM (optional): if you supply an OpenAI-compatible API key, summaries are generated by a chat model (default gpt-4o-mini).

The default run requires no API key and makes no paid LLM calls — it returns real summaries from the built-in extractive summarizer (the actor still fetches the RSS feeds you provide over the network).

Usage example

Minimal input (free extractive mode, no API key):

{
  "rssUrls": ["https://rss.nytimes.com/services/xml/rss/nyt/Technology.xml"],
  "summaryLength": "medium",
  "maxItemsPerFeed": 10
}

With LLM summaries (supply an OpenAI-compatible key as a secret input, or set the OPENAI_API_KEY environment variable):

{
  "rssUrls": ["https://feeds.arstechnica.com/arstechnica/index"],
  "summaryLength": "short",
  "maxItemsPerFeed": 5,
  "openaiApiKey": "sk-...",
  "model": "gpt-4o-mini",
  "openaiBaseUrl": "https://api.openai.com/v1"
}

Input

Field	Type	Default	Notes
rssUrls	string[]	NYT Technology feed	RSS/Atom feed URLs (http/https).
summaryLength	enum	medium	short (1 sentence), medium (3), long (6).
maxItemsPerFeed	integer	10	Cap per feed (1–100). Controls cost/time.
openaiApiKey	string (secret)	—	Optional. Enables LLM summaries. Can also use the OPENAI_API_KEY env var.
model	string	gpt-4o-mini	LLM model (when key is set).
openaiBaseUrl	string	https://api.openai.com/v1	OpenAI-compatible base URL.
requestTimeoutSecs	integer	30	Per-request timeout (1–120).

Output

Each dataset record:

{
  "feedUrl": "https://rss.nytimes.com/services/xml/rss/nyt/Technology.xml",
  "feedTitle": "NYT > Technology",
  "title": "Article title",
  "link": "https://www.nytimes.com/...",
  "pubDate": "Fri, 20 Jun 2026 12:00:00 +0000",
  "summary": "Concise summary text.",
  "summaryLength": "medium",
  "summaryMethod": "extractive",
  "scrapedAt": "2026-06-20T12:00:00.000Z"
}

summaryMethod is one of extractive, llm:<model>, or extractive (llm-fallback) (used when an LLM call fails so the run still produces output).

Cost & reliability notes

• Without an API key there is no LLM cost.
• With an API key, cost scales with feeds × maxItemsPerFeed. Keep maxItemsPerFeed low to bound spend.
• Feeds that fail to fetch, are empty, or are malformed are skipped with a warning; the run continues with the remaining feeds.
• If an LLM call fails (rate limit, bad key, timeout) the item falls back to the extractive summary instead of crashing the run.

Limitations

• The extractive summarizer selects the leading sentence(s) of each item's content; it does not paraphrase or abstract. Summary quality depends on the feed providing usable content:encoded/description text (some feeds give only titles, in which case the summary is derived from the title).
• LLM summaries require your own OpenAI-compatible API key; cost and rate limits are governed by your provider, not this actor.
• If every supplied feed fails to fetch/parse or has no items, the run reports a warning and pushes 0 records. If no valid http(s) feed URL is supplied at all, the run fails cleanly with an explanatory message.
• maxItemsPerFeed is capped at 100 and requestTimeoutSecs at 120, regardless of higher values passed.