Kendell

Numerous reliability improvements. Largest one: support for non-Chrome browsers! It works in Firefox per this screenshot, it works in Chromium, and it even works in Webkit (Safari/GNOME Web), all thanks to some unreleased Scramjet bug fixes. Also added a better hint that using 0K works best in Chrome, improved the URL parser, and better organized docs.

One more thing: if you don’t want to set up your own tunnel to test 0K at all, you can now run a temporary tunnel that pipes your traffic through a third party Wisp server at https://0k-web.github.io/relay/

It’s now actually easy to run a tunnel: instead of having to clone 0K and find the relevant code, you can just download a compiled binary, run npx @0k-web/server, or use an online sandbox as a tunnel.

I made a few bug fixes and I rendered an OG image. I’m not actually using OpenGraph, instead uploading it to Flavortown, GitHub, and my project index.

WebRTC time! 0K will now ask you for a tunnel code to access the rest of the internet. When you enter it, 0K makes an offer (it goes to Google AMP to a Cloudflare DO to wherever the tunnel is), listens for an answer, and establishes a connection to the tunnel. Each tunnel code is a single word - good for increasing your vocabulary, and surprisingly unlikely to collide (there’s 370k words).

If you don’t get the point of 0K, I wrote an explainer and added it to the README.

0K now builds to SVG. Yes, SVG. This lets CDNs like jsDelivr and Statically serve it. (Inspired by Night’s DayDreamX)

I took another detour from main development. I had to bring the size of the assets down so 0K can run anywhere HTML runs. I ended up taking advantage of jsDelivr for third party assets and tweaking my zip.js import to skip heavy WASM and content type lookup data. And of course, I added a GitHub Releases workflow so you can actually download 0K.

The promised transport update: local sites now are supported! These let you view a .zip, whether from GitHub or of your own sourcing, as a real webpage, You add local sites (or switch github.io routing between raw.githubusercontent.com, jsdelivr.net, or turned off) via the new settings panel.

I took a detour to work on the home page and README. Back to developing the transport after this.

It’s time to implement the part that makes 0K special: the transport that works without a server. So far, it supports loading GitHub Pages (github.io) via raw.githubusercontent.com, and loading a custom home page.

And I added to the controls: now you can go back/forward/reload. You can’t actually use back/forward on an iframe, and I didn’t want to make a technically incorrect implementation of back/forward, so I’m instead displaying the whole navigation history.

I added some controls. Styling them to make them feel smooth and unintrusive, and figuring out how to track the current URL, was hard but rewarding.

I have the core, Scramjet 2, working! Scramjet 2 actually works (although, fun fact, example.com doesn’t due to some SSL technicalities). I had to apply a few patches and write a few custom Vite plugins to handle Scramjet Vite-style (like bundling, inlining, and asset hashing).

Even just initializing the basic empty Vite setup, I had a few false starts. Vite didn’t let me inline styles and scripts so I had to use a plugin. That plugin didn’t inline the favicon but I decided the favicon was unnecessary. And the framework I want to use, Scramjet, didn’t have their version 2 ready (so I’m going to be stuck on version 1).

I redesigned the home page when logged out to include everything the README would, plus a designed header.

I finished developing Daily Bugs. Info about what it is is on the description section, the README, and on the website. Some of the most interesting highlights:

• This app is the first time I really took advantage of the domain I got recently; the Daily Bugs subdomain runs a Cloudflare Worker and Resend.
• I also took advantage of Cloudflare a lot. Aside from using a Cloudflare Worker with Cloudflare Static Assets, I also used KV for saving configuration and Workflows for every “detect bugs” run.
• Since I was working with the budget LLMs Hack Club AI offered, I had to do a lot of prompt engineering to get quality results. Most LLMs give you subjective code smells instead of objective bugs. Had to use LLMs to get this working to the point where some days it doesn’t even say “there’s bugs”.

Why only 1 devlog? I didn’t figure out how to fix Hackatime linking - by turning on “Allow public stats lookup” - until Daily Bugs was done.

I’m working on my first project! This is so exciting. I can’t wait to share more updates as I build.