mattx / milton Goto Github PK

Deleting an article does not remove its entry in the sidebar without a refresh, but it should.

MeiliSearch looks pretty cool. Written in rust, and open source. Not sure about the resource constraints, but maybe we can run it on a free instance or something. Lots of interesting features in the docs.

This will probably become more relevant once we start brushing up on the Algolia free tier limits.

The readability service should go through all relative links in the document and make them absolute.

We could delete when we reach a given downvote count

This would enable link sharing

Maybe make it accessible from Discord?

When a new article is submitted, check if an article with very high text similarity already exists, and if yes, don't index it.

This would be more robust than trying to deduplicate URLs manually.

Categories for article types

When a link is submitted to Milton, it would be nice if we automatically submitted it to a web archiver like this to ensure that we have a reference to the original content in the face of link rot.

It would be cool if we could submit links to PDF's, and get a "reader" view similar to what happens for regular HTML webpages right now. Existing Milton-esque tools like Pocket don't support this use case for PDFs, which is especially frustrating on mobile/ small screens, where reading PDF's often require lots of side to side scrolling.

There are various open source libraries, of which Mozilla's PDF.js seems like the most promising, but Apache Tika is also interesting since it supports a lot of formats via a single interface, which would be useful if we wanted to extend this functionality to other formats in the future (Microsoft Word documents come to mind).

For extra credit, it would be even cooler if Milton was smart enough to fetch the paper when given an Arxiv or other paper aggregator link (similar to what it does for HN or Reddit?)

When asked to index a comment page on an aggregator such as Reddit or Hacker News, we should index the thread the page is about, not the comment page.

We probably need to implement specialized logic for each aggregator.