Hi! I'm the dev here! I built this on a whim at after seeing someone ask for it on twitter. It was 12:30 at night but I couldn't pass down the opportunity to build it.
The code is very simple, there's no backend at all actually, I believe because wikipedia's api is very permissive and you can just make the requests in the frontend. So you just simply request random articles, get some snippets, and the image attached!
I used Claude and cursor do 90% of the heavy lifting, so I am positive there's plenty of room for optimizations. But right now as it stands, it's quite fun to play with, even without anything very sophisticated.
Shoutout to APIs that do not enforce CORS preventing requests be made from FE without a need for a BE. There's so many toy apps I started building that would have just worked if this was more common, but they have CORS restrictions requiring me to spin up a BE which for many one-off tools and personal tools just isn't worth doing and maintaining. Same with OAuth.
Yes, exactly. People who want to "disable" it have no idea how the web works. Developers have all kinds of misconceptions about what it is, I even heard someone saying it disallows backends to call their API.
And in particular CORS is the region you can read the wikkpedia api cross origin (unless you are doing jsonp, but hopefully they are using CORS because it is better in every way)
I would only use something like this that requires absolutely no authentication. For example, I had a one page app that showed me instantly when the next shuttle(s) were scheduled for my stop. Instead of having to click through multiple steps, this allowed me to see it in one step. As far as I know, I was the only user for this thing I built and put up on gitlab pages. I don't know exactly because I didn't bother to track who visited the page.
This is the way to go, you wouldn't want to use a CORS proxy for something authenticated/with credentials (e.g. API key). But for public unauthenticated request, they work just fine.
The only caveat I feel is that the speed of the API is definitely not comparable to something more purpose built for this kind of scale, but overall I'm happy as it works well enough that I don't have to think about it too hard.
I think Github Actions could be used for scheduled builds, so that the initial load would have random articles right in. Further requests could then be made in advance so users would not notice any delay from the API.
I attempted to implement the schedule trigger [1] on GitHub Actions as an example, but it is not being triggered as I expected. It needs more digging if you're so inclined.
Aside from that, the whole gist is that the initial data can be injected into the static files during the build step, or even saved as separate JSON files that the app can load instead of reaching out to the API. As long as you're willing to refresh the static data from time to time, of course.
I created a basic example at https://schedbuild.pages.dev/ with a rough, manual implementation of a build step. Frameworks like Next.js offer a more sophisticated approach that can render the entire HTML, allowing users to load the static page with the initial data already rendered without Javascript, and subsequent interactions taking over from there more seamlessly.
If the Github Actions schedule feature is ever sorted out, in my opinion it's a reasonable alternative to setting up a backend just for this.
Edit to the other comment: the cron job wasn't being triggered at first, but turns out it's just slightly delayed. The example has been updated hourly since then.
You can use a schedule trigger [1] on GitHub Actions.
The whole gist is that the initial data can be injected into the static files during the build step, or even saved as separate JSON files that the app can load instead of reaching out to the API. As long as you're willing to refresh the static data from time to time, of course.
I created a basic example at https://schedbuild.pages.dev/ with a rough, manual implementation of a build step. Frameworks like Next.js offer a more sophisticated approach that can render the entire HTML, allowing users to load the static page with the initial data already rendered without Javascript, and subsequent interactions taking over from there more seamlessly.
In my opinion this is a reasonable alternative to setting up a backend just for this.
Shameless plug for Magic Loops -- we run code in isolated MicroVMs and students love our lack of CORS enforcement, as the APIs they build can be easily integrated into their hackathon projects :)
This is awesome. I can imagine you likely are not interested in building one but this site could hugely benefit from a recommendations algorithm.
For example an algorithm could understand how much a user really enjoys a certain article and then starts sending the users down a rabbit hole of similar and tangential content. Designing, building and maintaining an algorithm like this though is no small feat.
I would not be surprised if Claude + Openai's reasoning models could develop a simple rudimentary algorithm that would work. Of course it wouldn't be as sophisticated as something like TikTok and would require a lot of fine tuning, but it's definitely possible.
Personalization is not inherently bad and I believe personal feeds can be engineered in a way that doesn’t result in echo chambers but in communities.
The problem is capital incentives are not aligned with making these interfaces which is why we have the feeds we have today on Facebook, Twitter, etc. I look forward to the innovations happening on bring-your-own-algo Bluesky.
As a side note, I’m currently building a personalized Hacker News service. I might throw it at Show HN once it gets closer to completion.
I think tomorrow I'm going to write a detailed blog of exactly what I did building out WikiTok. One last little bit of info on this subject (and at the end of it I'll say that I'm open to work or whatever).
As soon as I saw the tweet, I realized the opportunity was there waiting for me. And also, twitter's algorithm is REALLY good at pairing the right tweets to one another, so many people saw those two tweets side by side, which added to the humor.
Next steps: ingest these offline and process them into quick 30 second videos with the most salient facts. TTS narration, additional images. Generate stock video using the images in the API and perhaps text-to-video. That would be a killer app.
Bonus: come up with a heuristic or model to filter out or de-rank universally uninteresting articles.
> Indeed makes you wonder what's there about the sliding that makes it so entertaining.
Check out "skinner box" - the fact that you may get something interesting or may not is more exciting than just getting something good. They're lootboxes of information/entertainment.
This is freaking really cool, I’m at work browsing HN instead of doing actual work, but I’ll look into it more later, but the “killer feature” I think would be to add audio narration do this, or a quick summary, I would scroll that all day…
This is super super cool. I’ll tell you the single barrier to me actually using this regularly—it needs an algorithm. It would be so cool (and a good learning project) to even build the simplest of recommendation algorithms behind it based on my likes, dislikes, bookmarks, and whether I click “read more”.
Nice job whipping up something so simple, yet elegant, so fast.
This is what I love about LLM tools like cursor, it makes the effort to just try and build something so low, you can just try it one night, and can make cool things that might not have been built otherwise.
Very nice! I wonder if it would be possible to see articles related to a topic. Maybe using the hubs as a starting point and then following the related links in each article?
The code is very simple, there's no backend at all actually, I believe because wikipedia's api is very permissive and you can just make the requests in the frontend. So you just simply request random articles, get some snippets, and the image attached!
I used Claude and cursor do 90% of the heavy lifting, so I am positive there's plenty of room for optimizations. But right now as it stands, it's quite fun to play with, even without anything very sophisticated.
Here is the source code. https://github.com/IsaacGemal/wikitok