Endlessly Curious

Pintxøs: RSS without the clickbait

Featured image

Here is a headline that showed up in my feed reader recently:

Netflix’s Renewed Sci-Fi Thriller With Perfect Rotten Tomatoes Score Officially Hits A Filming Milestone

Which thriller? What milestone?

Here is what it actually says:

Netflix’s Supacell completes season 2 filming

That’s it. That is all the information in that “article”. The first version is written for one purpose only: to make you click.

I read most of my news through RSS (I mentioned NetNewsWire and Feedbin in the lens feed post a while ago). RSS is great precisely because there is no algorithm between me and the sites I chose to follow. I just get a simple list of the articles that are published. But headlines themselves have gotten steadily worse. You have to click to find out, which is of course the whole point. I understand why they do it. But that doesn’t make it any less annoying.

So I built Pintxøs. It runs at home in a Docker container. You give it the RSS feeds you already read, it fetches them, sends every article through Claude Haiku once to write a plain, factual headline and a short summary, and serves the result back as a clean RSS feed. You subscribe to that in your reader instead. One API call per article, cached, so a feed with fifty items a day costs cents at most.

If you read via an RSS reader, this might be for you. It runs anywhere you can run a Docker container.

Pintxos are the small, delicious bar snacks you get in the Basque country, a slice of bread with something on top, held together with a toothpick. I live in Spain; the association was hard to avoid. The tool takes an article and skewers it down to the one tasty little bite that matters. I liked the ø-spelling because it reminds me of Denmark and of a toothpick stuck straight through the o. Also, it’s silly. Works for me.

The Pintxøs feeds page: wordmark, three feeds in a hairline table, output URL chips with copy buttons, poll and delete actions

This tool has no tracking, no accounts, no commercial anything. The code is on GitHub and published under an MIT license, the container is one docker compose up away. But you need to bring your own Anthropic API key.

How I use it in practice:

Other candidate feeds are those that publish long articles with high frequency. Even if they are not full of clickbait, it can still help to get shorter, more factual headlines and summaries.

What’s next? Two things I have on my list:

  1. Support for different providers and models. Claude Haiku works very well. But Anthropic’s API costs are still quite high in general. I want to experiment with other, cheaper models.
  2. Adding keyword filters. Say you are subscribed to The Verge. Its feed contains a TON of articles every day. Maybe you already know that there are certain types of articles you just don’t want or need to see. It would be nice to filter out all articles containing certain key terms from the output feed Pintxøs provides.

So more to come.

Disclosure: This post was created together with Claude: the working session wrote up its own notes, Claude drafted the post from them and my input, and I edited it heavily.