Why I Built Sigilex

I've been into knives for a while now. The hobby kind, where you develop opinions about steel composition and grind geometry that you genuinely cannot explain to normal people without watching their eyes glaze over. There are a handful of makers I actually care about. Small shops, custom runs. They announce a new drop and it sells out in three hours if you weren't paying attention.

I was never paying attention.

Not because I didn't care. Because finding the signal was its own kind of labor. I'd check Reddit and get buried under beginner questions, memes, and the same "what's a good first knife under $50" thread for the hundredth time. By the time something actually relevant to me surfaced, the drop was gone.

Same thing with Warhammer. I'm specifically into the Horus Heresy. I don't need Age of Sigmar painting tutorials. I definitely don't need whatever the latest faction drama is. But that's what I was getting. And Warhammer has this whole limited-release thing going on, where Black Library puts out 1,500 signed copies of a special edition and they're on eBay for ten times retail within the hour. I don't even care about the signature. I just want the leather-bound book because I think it looks cool on the shelf. But you have to be fast, and the feeds are not set up to help you be fast.

Knives are the same. Pipes are the same. I've gotten into pipe tobacco recently, which has a surprising amount of depth to it if you're curious. Cigars too. And honestly, it's the same pattern in every one of these communities. There's stuff I'm really interested in, and it's surrounded by a volume of stuff I'm not. The labor of sorting through it is real, and nobody is doing it for me.

At some point I asked myself: is this actually saving me time, or am I just tolerating the situation?

The honest answer was the second one.

I think this goes back further than I initially realized. I keep coming back to Google Reader. For people who used it, Reader was probably the closest the internet ever got to "you decide what you see." You built your sources manually, everything was chronological, and it just worked. It was a reflection of your interests that didn't try to be clever about it.

Then Google killed it, because Google kills things.

After that there was Feedly, and Feedly was fine for a while. But whoever is running product over there decided at some point that this is an enterprise tool. The direction is clearly aimed at corporate security teams monitoring CVEs. Which, honestly, is a terrible solution for that. There are much sharper tools for it. But that pivot means Feedly isn't doing a great job for me when I want to learn about new Iron Warriors releases in the Black Library.

The deeper problem with RSS in general is that you have to know your sources ahead of time. There's a dimension of discoverability that's just missing. I'm not always going to know which subreddit or blog or forum has the thing I'll care about. I need something that can cast a wider net.

I should talk about X, because I think it's relevant.

I uninstalled it from my phone recently. I was getting some genuinely useful information from it, particularly around AI model releases. But after using it extensively for a couple of months, I noticed a pattern. Every single person there is trying to sell something. Engagement is the product, and when you pay people for engagement, they're going to do whatever it takes to optimize for it. Everything is hyperbolic. Everything is a take. A model release isn't "here's what changed." It's "new model just dropped, call your hypebeast friends."

The result is a feed that makes me feel worse when I walk away from it. I'd sometimes get useful content, but it was buried in hype, and the page itself was 80% stuff I didn't care about. Ads, algorithmic recommendations, engagement-optimized noise. It's a huge distraction and sensory overload.

I miss being able to view webpages that were just text. Just the content. That's partially why I work in a text editor and nothing else. I want a clean interface that optimizes for signal, not engagement.

There's a phrase for this that I think captures it exactly: the algorithmic attention economy. Your attention is the product. The algorithm's job is to keep you scrolling, not to show you what you care about. Those are fundamentally different objectives, and every major platform has chosen the first one.

Fuck that.

I mean it. My attention is mine. I get to decide what's worth it. And the fact that a handful of companies have built trillion-dollar businesses on the premise that they know better than I do what I should be looking at is something I'm not interested in tolerating anymore.

I run a pretty substantial homelab. Kubernetes cluster, dual-site, eight nodes across two physical locations, approaching 100 terabytes of spinning disk, probably 60 terabytes of NVMe. Declaratively controlled Talos clusters on top of Proxmox hypervisors. It's the kind of thing I do, often just for fun. I have two machines with dual RTX 5090s and 4090s running local LLM inference. Not as a novelty, but as real infrastructure.

I've been using language models seriously for long enough to have developed real opinions about where they're useful and where they're embarrassing. They're genuinely useful for classification. Give a model enough context about what you care about, and it can look at a piece of content and tell you whether it's relevant to you. Not perfectly. But well enough to be valuable. That's a narrow enough task that the failure modes are manageable.

I also run about 50 different open source projects in my cluster, and one of the things that takes real time is tracking software releases. I'm trying to follow version bumps across Longhorn, Flux, cert-manager, ingress-nginx, and dozens of others. The labor of checking each project, scanning the changelog, and mapping what changed onto what I actually care about is genuinely taxing. It's useful work, but it shouldn't require a human to do it manually.

So I started building something for myself. Not a product, just a tool. A pipeline that would ingest content from the sources I cared about, run it against a profile I wrote describing my interests, and deliver me a daily briefing with only the stuff that matched. Knife drops. Warhammer releases for the factions I play. CVE alerts for my actual stack. PRS match results.

The first time it worked, I sat there for a minute.

I'd been conditioned to expect a certain amount of noise and just accept it. This was different. It was just the stuff I cared about. Nothing else.

I want to be honest about something, because I think it matters.

There's a lot of AI fatigue right now, and I think most of it is earned. The industry has spent three years teaching people to associate "AI-powered" with "slightly worse version of a thing you already had, now with a chatbot." The products are often lazy. The copy is lazier. I think the whole category has started to feel like a costume people put on existing tools.

I'm building Sigilex with AI as a core component. So I've had to sit with that.

My conclusion is pretty simple. The fatigue is about how AI is being used, not AI itself. Most of these products are using language models as a shortcut to replace thinking about what the product should actually do. The model becomes load-bearing where product thinking should be.

What I'm doing is narrower. The AI in Sigilex does one thing: it classifies content against your interests, which you define in plain language. That's it. It doesn't summarize for you. It doesn't tell you what to think. It does the filtering work so that what reaches you has already passed through a lens you designed.

AI is an implementation detail of a useful product. It's not the feature. I think that distinction matters, and I think the market is going to sort pretty aggressively along that line.

I should tell you what I'm not building, because I think it clarifies the thing.

I'm not building another RSS reader. RSS readers are good, actually. But they're manual, they're technical, and they still deliver everything from a source instead of the things from a source you actually care about. The management overhead is real.

I'm not building a social platform. No likes, no followers, no engagement metrics. The system has no reason to care about anything except whether a piece of content is relevant to you.

I'm not building a generic news aggregator. "AI news summary" is a product category I find pretty uninteresting. Summarization is a convenience feature. Classification is the value.

The closest mental model for Sigilex, for me, is this: imagine you had a friend who reads everything you'd want to read, knows your actual interests in granular detail, and sends you a daily message with only the things you'd genuinely care about. That's the experience I'm building toward.

One daily briefing. Your topics. About ten minutes to read. And then you're done.

That's the part I think people underestimate. Every algorithmic feed is designed to be infinite. There is no bottom. You can't finish Twitter. You can't complete Reddit. The scroll goes on forever, and that's by design, because the longer you stay, the more ads you see. The anxiety you feel when you close the app isn't a bug. It's a feature of the attention economy.

Sigilex turns an infinite, anxiety-inducing feed into a finite, completable task. You open your briefing. You read the things that matter. You're caught up. You close it and walk away, and you feel better than you did before you opened it, not worse.

That's the whole product, really. That feeling.

I also want to say something about ownership, because it's important to me.

I care about owning my own things. I really don't like it when a company tells me what I should care about. The open source movement speaks to me because of this. Tools should be working for me, not the other way around.

Sigilex is built on a simple premise: the algorithm should be yours. You describe what you care about. The system works for you. Nobody else gets a vote on what shows up in your briefing. No ads. No engagement incentives. No algorithmic middleman deciding that actually, you'd probably prefer to see this trending topic instead of the thing you asked for.

In many ways, this whole thing is my answer to the attention economy. I'm taking my attention back. The platforms had it. I want it. That's the whole dispute.

The thing I'm building for myself is at sigilex.com. It's not ready for you yet, but it will be soon.

If you're the kind of person who cares deeply about specific topics and is tired of the algorithm deciding what you see, whether that's knife drops, limited edition books, CVE alerts, model releases, or match results for the sport you actually compete in, I think you'll get it.

Thanks for reading.

— Erik, Ashfall