From the readme:
This library (including the schema documentation) was largely written with the help of Claude, the AI model by Anthropic. Claude's output was thoroughly reviewed by Cloudflare engineers with careful attention paid to security and compliance with standards. Many improvements were made on the initial output, mostly again by prompting Claude (and reviewing the results). Check out the commit history to see how Claude was prompted and what code it produced.
"NOOOOOOOO!!!! You can't just use an LLM to write an auth library!"
"haha gpus go brrr"
In all seriousness, two months ago (January 2025), I (@kentonv) would have agreed. I was an AI skeptic. I thoughts LLMs were glorified Markov chain generators that didn't actually understand code and couldn't produce anything novel. I started this project on a lark, fully expecting the AI to produce terrible code for me to laugh at. And then, uh... the code actually looked pretty good. Not perfect, but I just told the AI to fix things, and it did. I was shocked.
To emphasize, this is not "vibe coded". Every line was thoroughly reviewed and cross-referenced with relevant RFCs, by security experts with previous experience with those RFCs. I was trying to validate my skepticism. I ended up proving myself wrong.
Again, please check out the commit history -- especially early commits -- to understand how this went.
You can use it to produce any kind of multimedia webpage. Some restaurants are utilizing it by printing out the QR code it produces so customers can see their menu.
That was something I went back and forth on and eventually decided that there was too much negative space when the browser chrome disappeared but all the content is accessible, or should be anyway. If not please let me know. Thanks.
There’s two sides to it, an author creating something using plain text in the authoring tool and a reader viewing the final product. So the author uses plain text to get their point across to the reader.
When you’ve created something you can publish it and the reader can view it at a URL. The author obviously isn’t trying to get a point across to themselves in the app.
The authoring tool is a web app and does quite a lot so it is JS heavy. It works well on mobile but the demo only works in landscape as the mobile version is slimmed down so you do have to rotate your phone.
In this case I meant you can use plain old text to render a webpage… which is what you do anyway but I thought I’d make up another way that people who love text but don’t code might find useful.
That was something I thought deeply about. I decided that an imaginary user would see a page change as a navigation event. Pressing back and having the whole thing poof could be just as anger inducing. Most SPAs use some type of router to do just that.. I imagine I won’t hear from anyone who prefers it that way so until I can afford some user testing I won’t know. Quite a bit of anxiety about that one. I had initially used replaceState…
Yes, plain text to create it not displayed as plain text. I didn’t write the post so the title may be a little misleading. Flashy JS heavy output for minimal input is the point.
"NOOOOOOOO!!!! You can't just use an LLM to write an auth library!"
"haha gpus go brrr"
In all seriousness, two months ago (January 2025), I (@kentonv) would have agreed. I was an AI skeptic. I thoughts LLMs were glorified Markov chain generators that didn't actually understand code and couldn't produce anything novel. I started this project on a lark, fully expecting the AI to produce terrible code for me to laugh at. And then, uh... the code actually looked pretty good. Not perfect, but I just told the AI to fix things, and it did. I was shocked.
To emphasize, this is not "vibe coded". Every line was thoroughly reviewed and cross-referenced with relevant RFCs, by security experts with previous experience with those RFCs. I was trying to validate my skepticism. I ended up proving myself wrong.
Again, please check out the commit history -- especially early commits -- to understand how this went.