Alex (@OxAlexxx) header
Alex (@OxAlexxx) profile picture

Entrepreneur • Software Engineer 🚀 Building startups & failing forward 💡 Sharing lessons, tools & experiments 🧠 From 0 to 20k MRR — live.

Lisbon
Joined January 2024
Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)Photo by Alex (@OxAlexxx)
Let me introduce myself: 👨🏻‍💻 More than 10 years of professional experience in development 🤖 Main languages: php, python, go 💵 Launched couple projects with a total MRR of $50 📉 Funded a startup that closed after six months with a loss of $100k. 🇵🇹 I live in Lisbon for 2 years
5
3
54
3,001
732 bytes. Root on any Linux server built since 2017. No race conditions. No per-distributor offsets. No crashing the machine. One straight-line logic flaw that lived in the kernel for nine years. CVE-2026-31431, codenamed Copy Fail, lets an unprivileged local user write four controlled bytes into the page cache of any readable file. The same 732-byte Python script roots Ubuntu, RHEL, Amazon Linux, and SUSE without modification. The page cache is shared across the host, so a container escapes nodes. A CI pull request roots the runner. A shared dev server gives every user root. The write hits memory, not disk. No inotify, no forensic trace after cache eviction. Execve sees the corrupted binary and hands over a root shell. Patch to mainline commit a664bf3d603d or blacklist the algif_aead module. If you run multi-tenant Linux, Kubernetes, or self-hosted CI, assume you're exposed until you verify. Full PoC and technical deep-dive: copy.fail This was AI-assisted discovery. Xint Code scanned the crypto subsystem in an hour and pulled out the highest-severity finding.
178
«How is NoVPS better than Railway? Why should I trust an unknown project when there’s a major player out there?» We have managed databases that store backups separately, so in the event of such issues, you can save your project. We also have WAL streaming archiving, which allows us to restore the database even if you delete the backups.
83
Want to make your UX absolutely terrible? Let me show you how oh-my-zsh nails it. 👇
1
66
Result? I have to manually type my command again. Every. Single. Day. Beautiful colors are not good UX when you keep breaking my flow.
1
24
Moral of the story: Don’t ask users to make decisions during critical moments. Especially not daily. Especially not when they’re in the middle of something.
18
I've never understood how Reddit works: in a bunch of specialized communities, you can't post anything unless you've commented on posts in those communities. But if I have a problem and want to ask for advice, I can't because of the filters. In other words, the moderators of these communities are forcing users to go and spam other communities
46
Railway is down, deploys don't work. So it's time to try out novps.io, right?
42
It's really cool that you've implemented AI, but maybe you should have fixed the app first so it wouldn't have to be restarted every 10 minutes just to keep everything running? @k8slens
34
My first rejection from Y Combinator. I'm proud of it:
3
94
It seems that self-hosting CI/CD isn't such a bad idea after all...
28
Github Actions are down for the second time today. This is no longer funny, Microsoft.
106
Me: mom, can we buy AWS? I need a cloud for my startup Mom: no, we have an AWS at home AWS at home:
1
2
43
Sitting in a coworking space 40 minutes from home, praying that my home server doesn't drop the VPN connection. How did I end up here? 👇🏻
1
1
67
This isn't self-hosting anymore — it's self-troubleshooting. And sure, I knew what I was getting into. The savings are real (especially with my 20TB media server). But people launching actual products on their own hardware? Why would they put themselves through this?
1
15
The thing is — problems don't show up on day one. They hit you like this: you're away from home, everything crashes, and you're just praying SSH still works.
9