RSS for AI agents: give your LLM fresh web data via API

An AI agent that scrapes the live web on every run is slow, expensive, and fragile. A better pattern is to let a feed engine do the crawling once, then hand the agent a clean list of new items when it needs them. RSS was designed for exactly that.

Why RSS is a good interface for agents

• Structured: title, link, date, description — no layout to parse.
• Incremental: the agent only sees what changed since the last run.
• Cheap: one API call instead of a browser session per source.
• Auditable: every item has a URL and a timestamp.
• Composable: mix news, government notices, YouTube channels, and search results in one list.

The web becomes a queue of events instead of a pile of pages.

A minimal agent loop

1. Create feeds for the sources the agent cares about.
2. On each run, fetch the newest items from the API.
3. Filter by keywords, recency, or source.
4. Pass only those items to the model.
5. Store the last-seen timestamp or item IDs to avoid repeats.

This is the same pattern as a human reading a feed reader, just automated.

Working with the RSSFlux API

RSSFlux exposes a REST API under https://rssflux.io/api/v1. Authenticate with a Bearer token:

curl -H "Authorization: Bearer rss_YOUR_KEY" \
  "https://rssflux.io/api/v1/feeds"

Create a feed from any public page:

curl -X POST "https://rssflux.io/api/v1/feeds" \
  -H "Authorization: Bearer rss_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com/news","title":"Example News"}'

Read the newest items:

curl -H "Authorization: Bearer rss_YOUR_KEY" \
  "https://rssflux.io/api/v1/feeds/123/items?limit=50"

Items come back newest-first, with title, link, publish date, description, and thumbnail. That is everything an agent needs to decide what to read.

The same API also covers:

• Search across stored items
• Link Drop for saving URLs into lists
• Fluxions for combining sources into curated feeds
• xLikeWriter for publishing short updates to a live channel

Practical tips

Filter before you spend tokens. Send the model titles and descriptions, not full pages. Ask it to select the handful worth fetching.

Use dates, not vibes. Store the newest pub_date you have seen and only process items after it. This keeps runs idempotent.

Deduplicate aggressively. The same article can appear on multiple sites. Normalize URLs, strip tracking parameters, and cluster by title similarity.

Separate ingestion from reasoning. A feed engine's job is to fetch reliably; the agent's job is to reason. Mixing them makes both worse.

Respect rate limits. The Hobby + API plan includes 30 requests per minute and two keys. Cache responses and batch work instead of polling in a tight loop.

A note on freshness

No feed is real-time. RSSFlux refreshes on a schedule: 12 hours on Free, 45 minutes on Hobby, 15 minutes on Developer and Business. If your agent needs minute-level freshness, run it on a matching plan or supplement the feed with a targeted fetch.

Pricing and access

API access is included on Hobby + API ($13/mo) with 30 requests per minute and two keys, and on Developer ($19/mo) and Business ($80/mo), which also raise feed limits and refresh speed. See pricing for the full comparison.

Related reading: How to create an RSS feed for a website that doesn't have one and Automate a daily email digest from any RSS feed.