Mechanics
How the AI RSS digest actually works
How the AI RSS digest works: one snapshot of eleven feeds, filtering that runs in your browser, shareable filter links, and OPML export for any reader.
The pipeline, in one picture
A feed is a file. The site owner reads a published set of feeds on a schedule, keeps only the smallest useful fields — headline, link, source, publication time — and stores that as a snapshot. When you open the digest tool, your browser fetches that snapshot once and does all the filtering locally. Nothing you type is posted back.
That is why the page feels instant and why there is no sign-up: the expensive part (fetching eleven feeds) happens once for everyone, not once per visitor.
Why not fetch the feeds live in your browser
It was the first thing tried. Browsers block cross-origin reads unless the feed's server allows them, and most news sites do not — so a live version has to route every request through a proxy service. In testing, the free proxies returned rate-limit errors, demanded API keys, or were simply down. A digest tool that returns nothing two times out of five is not a tool.
The trade-off is freshness: the pool can be a few hours old, occasionally a day. The snapshot timestamp is printed above every digest, so you always know what you are reading. For "what happened in AI this week", that is fine.
Filter links: the closest thing to a subscription
The tool writes your filter into the page URL — keywords, window, sources, item count. Two things follow from that:
- Bookmarking is subscribing. Save
${your filter}to the bookmarks bar and tomorrow morning it opens pre-filtered. No email, no permission prompt, no unread badge. - Sharing is a link. Paste it in a team channel and colleagues see the same slice. Unlike a screenshot, it keeps updating.
Export formats
| Button | What you get | Best for |
|---|---|---|
| Copy as Markdown | Headlines grouped by source, dated, as links | Notion, Slack, a GitHub comment, a weekly roundup |
| Download .md | The same list as a file | Adding to a notes vault on a schedule |
| All AI feeds as OPML | Eleven feed URLs grouped by type | One-step import into Feedly, Inoreader, NetNewsWire, Reeder, FreshRSS |
What counts as a match
Matching is a case-insensitive substring test against the headline. Comma separates several words and any one of them is enough. That means gpt matches "GPT-6.2" and "chatgpt", while gpt (with a trailing space) would not match a headline where it appears mid-word. It also means you can exclude nothing: there is no negative keyword yet, so a noisy term is better removed from the box than filtered around.
Where the defaults come from
The three sources switched on by default — Hacker News, Hugging Face and The Verge — are there because they cover three different questions: what builders are excited about, what shipped, and what it means commercially. The AI RSS feeds page lists all eleven, with a note on what each one is like to read.
That is the whole mechanism
No account, no email, no queue to manage — one filtered page.