SRead

Inspiration

SRead started with a simple question: what if scattered Markdown could feel less like a folder of files and more like a living library? The knowledge behind it was built through long-running Codex /goal sessions. Smaller models were delegated focused research tasks over time, collecting, comparing, and organizing material into a coherent foundation. SRead is the reading experience that grew out of that process.

What it does

SRead turns dense Markdown knowledge into a calm, book-like reading experience. It provides structured navigation, full-text search, deep links, reading progress, bookmarks, reading preferences, and a mobile-first layout designed for uninterrupted reading. It makes a large knowledge base feel approachable, navigable, and genuinely worth returning to.

How we built it

We used Codex /goal as a long-running research and execution loop. Instead of asking one model to do everything at once, we let smaller models handle focused passes: gathering references, exploring topics, identifying patterns, validating structure, and feeding their findings back into the evolving knowledge base. The product itself is built with React, TypeScript, and Vite. Content is processed at build time, safely structured, indexed, split into lightweight reading units, and served as a fast static application. The result is deterministic, lightweight, and easy to deploy.

Challenges we ran into

The hard part was not collecting information. It was turning a constantly growing stream of research into something coherent. We had to preserve structure across thousands of ideas, keep navigation and search consistent, handle long documents without slowing down the browser, and make the experience feel editorial rather than mechanical. Coordinating long-running goals, model outputs, content quality, and product design was a challenge in itself.

Accomplishments that we're proud of

We are proud that SRead became more than a content dump or an AI-generated archive. It became a real knowledge-reading system. The research is organized, the reading flow is intentional, and the interface stays quiet enough for the content to take center stage. We managed to combine deep research, automated assistance, careful information architecture, and a polished reading experience into one cohesive product.

What we learned

We learned that building a knowledge base is not just about finding more information. It is about creating relationships between ideas, removing noise, and giving people a reliable way back into what they have already discovered. We also learned that smaller models can be remarkably effective when given focused goals, enough time, and a clear pipeline for synthesis. The best results came from orchestration, iteration, and refinement—not from trying to solve everything in a single prompt.

What's next for SRead

SRead’s next chapter is about making the knowledge base even more useful: better discovery, richer connections between topics, more refined reading assistance, and smoother ways to turn long-running research into lasting understanding. The long-term vision is simple: use Codex and goal-driven model collaboration to continuously grow the knowledge, while keeping the final experience human, focused, and beautifully readable.

Built With

• accessibility
• codex
• content-pipeline
• full-text-search
• information-architecture
• knowledge-base
• local-first
• markdown
• offline-first
• openai
• progressive-web-app
• react
• responsive-web-design
• service-worker
• static-site
• tailwind-css
• typescript
• vercel
• vite