Building READER: From Information Overload to My Own AI Companion

Tired of expensive subscriptions and data tracking? Learn how I built a personal news reader with AI assistance as part of my fIOx system.

It’s late, as always, I’ve just finished a long workday, and I’m sitting on the train going home and staring at twenty open browser tabs. I have newsletters to catch up on, YouTube tutorials to watch, and tech blogs that I keep promising myself I’ll read “later.” But “later” never comes because my reading list has become a chaotic graveyard of forgotten links. Every time I find something interesting, I save it, and every time I save it, I forget to read it within the next few days.

For years, I tried to tame this chaos with traditional RSS readers—those tools that pull in news feeds so you don’t have to visit every website individually. I tried everything. Eventually, I was paying around 100 Euros a year for a premium subscription, only to find they were starting to charge extra for the very AI features I actually wanted. That was the breaking point. I realized that instead of paying someone else to gatekeep my information and track my reading habits, I could just build my own.

As part of my ongoing project, “Building fIOx”—the personal, 24/7 system I’ve been piecing together on a Mac Mini—I decided to create the “Reader.”

The Concept: A Library That Thinks With You

The concept is simple yet ambitious: a private, smart news reader and digital library that organizes everything I want to consume and uses AI to prepare it for me. It isn’t just a list of links; it’s a central hub for everything from long-form articles and e-books to YouTube videos and even my weekly digital newspaper.

The goal wasn’t just to have a place to store things, but to have a place that works for me while I sleep. It’s about turning a mountain of unorganized data into a curated, digestible stream of knowledge. Instead of me going to the content, the content comes to me in the format that fits my current situation. If I’m on a train, I want a podcast; if I’m on my e-ink tablet in the quiet garden, I want a clean article; if I’m at my desk, I want a deep dive. And thanks to Tailscale VPN, I don’t necessarily have to be connected to my home Wi-Fi, and can use it while traveling abroad.

A Workspace for Every Mood

One of the biggest frustrations with existing tools is that they force you into a single way of looking at information. They give you a list, and you have to deal with that list. I wanted more. I wanted my library to be as flexible as my attention span.

I spent a lot of time designing what I call Flexible Article Views. Depending on my mood, I can switch how I browse my content instantly. If I’m feeling overwhelmed, I switch to a simple, clean list of headlines to scan. If I’m in a visual mood, I can switch to a grid of beautiful cards or even a “video wall” that shows me thumbnails of all the YouTube content I’ve saved. For long browsing sessions, I use a magazine-style layout that feels more like reading a high-end publication than a tech feed.

To keep this from becoming a mess, I implemented Smart Grouping and Advanced Tagging. I don’t just dump everything into one bucket. I group my feeds into categories like “Technology,” “Smart Home,” or “Local News.” But the real power lies in the tagging. Unlike simple bookmarking, my tagging system allows for granular control. I can label an article with multiple specific tags, making it incredibly easy to find that one specific piece of information from three months ago without scrolling through hundreds of unrelated posts.

The Intelligence Behind the Interface

This is where the project moves from a simple organizational tool to a true assistant. I call it the AI Autopilot.

The most practical part of this is my custom “Later” Library. Most people have a “read later” list that eventually becomes a “never read” list. To solve this, I categorized my saved items by intent. I don’t just save things; I assign them to “To Read,” “To Listen,” “To Watch,” or “To Explore.” This way, when I have twenty minutes of downtime, I don’t have to decide what to do; I just look at my “To Listen” folder and let the system take over.

The real magic, however, happens in the background. While I’m working or sleeping, the AI Autopilot is busy. It can take a messy web page full of distracting ads, menus, and pop-ups and perform Seamless Content Extraction, leaving me with just the clean, high-quality text. It can transcribe audio files or generate summaries so I don’t have to read a 3,000-word article just to get the core idea.

And then, there is the feature I am most proud of: Custom Podcast Generation. I can take a group of articles I’ve marked as interesting and, through automation, have the AI bundle them into a single, custom-made podcast episode. It takes the text, audio or videos, processes it, and turns it into a spoken briefing. I can even choose the voice and the style of the summary. It has completely transformed my morning routine; I no longer wake up to a pile of unread emails, but to a personalized morning briefing that I can listen to while I’m making coffee.

My Reading Assistant

And whenever I read an article, my reading assistant is always there to help me—thanks to its connection to my local large language model via Ollama, it can perform all kinds of tasks. I’ve created several “AI Snippets” that I can trigger with a single click. For example, I can ask the AI to analyze the “logic” or numbers of an article and find any missing information or gaps in the argument, to generate a timeline of events mentioned in a news story, to perform a “Historian’s View,” simulating how someone might look back at this article in 10 years, to estimate if the article is really worth reading, to create a quiz based on the text to help me actually learn the material, and even act as a language tutor, explaining difficult terms or translating sections into English or German.

Another capability is the ability to have specific terms or phrasing explained within the article. By simply highlighting a word or entering a term manually, the AI uses the context of the entire piece to provide a detailed explanation.

Deep Interaction via fIOx

Because of the way I designed fIOx (if you missed the overview article of this series, it explains how the central API and MCP work), the Reader isn’t an isolated island. Every article I save is instantly accessible through my central system. This means I can use my Large Language Models (the AI engines that power my smart assistant) to interact with my articles in ways I never thought possible and even have a conversation with them.

What’s Next?

The Reader is a living system. It’s currently getting better at handling video content and improving the “Live Stream” feature via Voxtral, which allows me to have articles read to me instantly without waiting for a full generation. I’m also working on deeper integration for my e-ink tablet or my television, so my entire digital life—from YouTube to e-books—feels like one seamless experience.

This project has taught me that you don’t need to be a professional developer to build tools that solve your own problems. You just need a clear idea, the right AI assistance, and a bit of patience.