The current corpus of plaintext markup is over 150 GB with average HTML page load time of 81ms. Hosting images is much doable but would explode the storage and moderation requirements for self-hosters so it needs consideration.

There's a 3rd-party Telegram bot that bypass paywalls and returns a Nonograph URL with article contents. It works reasonably well and is quite fast, but it uses workarounds like scraping the homepage thousands of times per day to generate new anti-csrf tokens.

Something like it but first-party and built into the server to avoid trust in Telegram seems like a good start. After that comes archiving the full page, CSS, JS, and all.

It is depressing to see Nonograph long-term project get less interest than skills.md vapor-ware but such is life. <3 from Jiangmen China & Canada