Posts

Regular Movistar TV channels travel over Multicast/UDP, but everything you ask for yourself (a movie from the video store, rewinding what is currently on, a cloud recording) travels over Unicast, and that is where a home-made Linux router chokes. In this post I explain how that traffic works today, measured with the TV switched on, and what you need to put on the router to make it play. There is a surprise: not everything uses RTSP anymore. More than a decade ago I wrote Video on Demand for Movistar, the first time I wrestled with this on my Linux router. Recently I felt like checking whether igmpproxy and the kernel RTSP helper were still needed, I took another look at the whole thing, and this post is the result. Yes, I was bored and decided to do an academic study. …

A few weeks ago I shelved my “homemade Bitwarden” and moved to 1Password on the family plan. Yes, you have to pay, but sharing with the family and the ease of use make up for it. Shortly after migrating the passwords I ran into the Developer tab and its SSH agent: private keys live in the vault, they sync themselves across my machines, and every use is authorized with a fingerprint or with Windows Hello. By the way, I later found out that Bitwarden has had an SSH agent too since early 2025. You learn something new every day. It sounded too good to be true, so I set out to check whether it fit my case: several workstations (macOS, Windows, Linux), lots of servers, network gear with SSH from another era, and more and more scripts and AI agents running ssh without me being at the keyboard. In this post I go through what I learned, how I set it up, and where I do NOT use it. …

Sancho is skeptical by design, and with Skills you need to be more skeptical than ever. In my note “Sancho Learns Skills” I described how convenient it is to download a ready-made Skill and plug it into your agent. What I didn’t mention is the dark side: every Skill you download from a public marketplace is someone else’s code that your agent will run with your permissions. That’s the new weak link in the supply chain: attackers are already poisoning these repositories with malicious Skills. In this note I lay out a few ideas to protect yourself. …

When you work with an AI harness (Claude Code, Gemini CLI, Codex, …) you end up spinning up several instances over and over. I wanted a keyboard shortcut that opened my most typical setup with multiple panes in a single window. I called it the terminal “AI mode”: four panes, three running claude (each with a different model: opus, sonnet, haiku) and a fourth with a clean shell for auxiliary commands. The non-negotiable requirement is that all four panes start in the directory I pressed the shortcut from, with zero manual steps. This post covers two paths to set it up: WezTerm (recommended, cross-platform — Windows / macOS / Linux) and iTerm2 (fallback for those who already live in iTerm on macOS and don’t want to switch terminals). …

If you use AI agents or do vibe coding, you know how quickly repositories pile up across different accounts and providers. Personal GitHub, work GitHub, or even a self-hosted server like Forgejo… and suddenly you have dozens of repos cloned with the wrong identity, crossed credentials, and a monumental mess. This happened to someone very close to me — a non-technical person who is creating many “projects” with agentic AI and, like anyone starting with Git for versioning and backups, was going crazy. I created Gitbox to help “non-developers” manage multiple accounts and clones more easily, without them stepping on each other. If you’re a developer, think of it as another alternative — and if you like it, go ahead. I use it on all my desktops and headless machines. …

Downloading LLM models locally is something you do occasionally, but when you pull a massive 122B parameter model like Sehyo/Qwen3.5-122B-A10B-NVFP4, the download hogs the entire connection and leaves the rest of the household without internet. The Hugging Face CLI (huggingface-cli or hf) doesn’t have a --limit-rate flag, so you need to find alternatives. In this post I explain two ways to limit bandwidth on Linux using Docker (my preferred method) or Wondershaper at the host level. …