Key Facts
- ✓ The tool uses Gemini 3 Flash to generate HTML and Canvas posts just-in-time.
- ✓ Comments and DMs are linked via Cloudflare Workers Durable Objects for speed.
- ✓ Generated posts are saved into DO SQLite for faster serving in the 'Following' feed.
- ✓ The project was inspired by Wikitok, a VSCode extension created by the developer.
Quick Summary
A new experimental interface has been introduced that mimics the scrolling behavior of social media applications like TikTok but applies it to Wikipedia content. The project utilizes fully generative UIs, meaning the visual structure of the feed is created dynamically rather than being pre-loaded.
The system generates posts just-in-time using HTML and Canvas elements. This generation is powered by Gemini 3 Flash, which streams the content to the user. The developer has also integrated Cloudflare technology to handle the speed of interactions, specifically for comments and direct messages.
This project was inspired by previous work, including a VSCode extension called Wikitok. The goal is to provide a faster, more engaging way to consume information by serving saved posts quickly through a specific database architecture.
Generative UI Architecture
The core of this new experience is the generative UI technology. Unlike traditional websites that serve static HTML, this tool generates the visual layout of every post on the feed in real-time. The developer has experimented with this concept since the era of GPT-2, but the current implementation uses Gemini 3 Flash to handle the heavy lifting.
Every post seen on the feed—ranging from various topics—is streamed directly to the user. The system creates the HTML and Canvas elements just moments before they appear on screen. This approach allows for a highly dynamic and potentially infinite stream of content.
The technical stack involves:
- Real-time generation of visual elements
- Streaming via Gemini 3 Flash
- Dynamic HTML and Canvas rendering
Speed and Data Management
Speed is a critical factor in the user experience of this tool. To achieve responsiveness similar to native applications, the developer implemented Cloudflare Workers Durable Objects. This technology manages the state for comments and direct messages, ensuring they feel instantaneous and bidirectionally linked.
Behind the scenes, the system handles data storage efficiently. Every generated post is saved into a DO SQLite database. This storage method allows the system to serve content rapidly to the 'Following' feed.
By caching the generated posts, the application can deliver a smoother scrolling experience. The combination of Cloudflare infrastructure and local storage optimization aims to eliminate lag when switching between posts.
Inspiration and Origins
The project did not emerge in a vacuum; it is the result of iterative experimentation. The developer cites Wikitok as a primary source of inspiration. Wikitok was originally created as a VSCode Extension focused on the concept of brainrot—a term often used to describe highly addictive, short-form content.
In addition to the VSCode extension, the developer had previously built another fully generative UI site. These prior projects provided the technical foundation and conceptual framework needed to build the current Wikipedia-scrolling tool.
The evolution from a code editor extension to a standalone web interface demonstrates a shift toward making information consumption more accessible and visually engaging for a broader audience.
Conclusion
This experimental tool represents a significant shift in how information from sources like Wikipedia can be consumed. By combining generative AI with the interface patterns of social media, the developer creates a unique browsing experience. The use of Cloudflare and DO SQLite ensures that the experience remains fast and responsive, despite the complex real-time generation happening in the background.
As users continue to look for more engaging ways to learn, tools that blend education with entertainment mechanics may become more prevalent. This project serves as a proof-of-concept for how just-in-time generation can be applied to large knowledge bases.
