Why I love NixOS

What I love about NixOS has less to do with Linux and more to do with the Nix package manager¹.

To me, NixOS is the operating system artifact of a much more important idea: a deterministic and reproducible functional package manager. That is the core of why I love NixOS. It is not distro branding that I care about. It is the fact that I can construct a whole operating system as a deterministic result of feeding Nix DSL to Nix and then rebuild it, change it bit by bit, and roll it back if I do not like the result.

I love NixOS because most operating systems slowly turn into a pile of state. You install packages, tweak settings, try random tools, remove some of them, upgrade over time and after a while you have a machine that works but not in a way that you can confidently explain from first principles. NixOS felt very different to me. I do not have to trust a pile of state. I can define a system and build it.

I love NixOS because I can specify the whole OS including the packages I need and the configuration in one declarative setup. That one place aspect matters to me more than it might sound at first. I do not have to chase package choices in one place, desktop settings in another place and keyboard behavior somewhere else. Below are a couple of small Nix DSL examples.

• GNOME extensions:

environment.systemPackages = with pkgs; [
  gnomeExtensions.dash-to-dock
  gnomeExtensions.unite
  gnomeExtensions.appindicator
  libappindicator
];

services.desktopManager.gnome.extraGSettingsOverrides = ''
  [org.gnome.shell]
  enabled-extensions=['dash-to-dock@gnome-shell-extensions.gcampax.github.com', 'unite@hardpixel.eu', 'appindicatorsupport@rgcjonas.gmail.com']

  [org.gnome.shell.extensions.dash-to-dock]
  dock-position='BOTTOM'
  autohide=true
  dock-fixed=false
  extend-height=false
  transparency-mode='FIX'
'';

• Key mapping per keyboard:

services.keyd = {
  enable = true;

  keyboards = {
    usb_keyboard = {
      ids = [ "usb:kb" ];
      settings.main = {
        leftcontrol = "leftmeta";
        leftmeta = "leftcontrol";
        rightalt = "rightmeta";
        rightmeta = "rightalt";
      };
    };

    laptop_keyboard = {
      ids = [ "laptop:kb" ];
      settings.main = swapLeftAltLeftControl;
    };
  };
};

Those are ordinary details of a working machine, but that is exactly the point. I can describe them declaratively, rebuild the system and keep moving. If I buy a new computer, I do not have to remember a long chain of manual setup steps or half-baked scripts scattered all over. I can rebuild the system from a single source of truth.

I love NixOS because it has been around for a long time. In my experience, it has been very stable. It has a predictable release cadence every six months. I can set it up to update automatically and upgrade it without the usual fear that tends to come with operating system upgrades. I do not have to think much about upgrade prompts, desktop notifications or random system drift in the background. It mostly stays out of my way. And if I want to be more adventurous, it also has an unstable channel² that I can enable to experiment and get newer software.

I love NixOS because it lets my laptop be boring in the best possible sense. I recently bought an HP laptop³ and NixOS worked beautifully on it out of the box. I did not have to fight the hardware to get to a reasonable baseline. That gave me exactly what I want from a personal computer: a stable system that I can configure declaratively and then mostly ignore while I focus on actual work.

I love NixOS because it makes experimentation cheap and safe. I can try packages without mutating the base system. I can construct a completely isolated package shell⁴ for anything from a one-off script to a full-blown project. If I want to harden it further, I can use the Nix DSL to specify the dependencies, build steps and resulting artifacts declaratively. That is a much better way to work than slowly polluting my daily driver and hoping I can reconstruct what I did later.

I love NixOS because I can use the same package manager across macOS and Linux. There is also community-maintained support for FreeBSD, though I have not used it personally. That is a huge practical benefit because my development tooling and dependency management can stay mostly uniform across those systems. It means the value of Nix is not tied only to NixOS. NixOS happens to be the most complete expression of it, but the underlying model is useful to me across platforms.

I love NixOS because it fits especially well with the way I work in the current LLM coding era.

Tools are changing very quickly. Coding agents often need very specific versions of utilities, compilers and runtimes. They need to install something, use it, throw it away, try another version and keep going without turning my PC into a garbage dump of conflicting state. Nix fits that model naturally. If I tell a coding agent that I use Nix, it is usually clever enough to reach for nix shell or nix develop to bring the needed tool into an isolated environment and execute it there. That is especially handy because Nix treats tooling as a declared input instead of an accidental side effect on the system.

A concrete example: I recently built a voice-to-text agent in Rust⁵. I did not have the Rust toolchain installed on my system. I simply told the coding agent that I use Nix, and it figured out how to pull in the entire Rust toolchain through Nix, compile the project inside an isolated shell and produce a working binary. My base system was never touched. No ~/.cargo, no ~/.rustup, no mutated PATH entries left behind. Without Nix, the agent would have reached for curl | sh to install rustup, quietly mutated my environment and left my system slightly different forever. With Nix, none of that happened.

This pattern generalizes. Every time an agent needs Python 3.11 vs 3.12, a specific version of ffmpeg, an obscure CLI tool or a particular compiler, Nix gives it a clean and reversible way to get exactly what it needs. The agent does not have to guess whether a tool is already installed or in the wrong version. It just declares what it needs and Nix takes care of the rest in a sandboxed way.

The other thing I appreciate is that Nix turns an agent's experiment into something you can actually commit and reproduce. Once the agent has a working setup, you can capture the exact dependencies in a flake.nix and run nix flake check to verify it builds cleanly from scratch. That transforms an ad hoc agent session into a reproducible, verifiable artifact. That is a much stronger foundation for delivering something that works reliably in production than hoping the environment happens to be in the right shape on the next machine.

I love NixOS because I like what Nix gives me in deployment too. I have never been a big fan of Docker as the final answer to the "works on my machine" problem. It solved important problems for the industry, no doubt about that, but I always found the overall model less satisfying than a truly deterministic one. Nix gives me a much better story. I can use dockerTools.buildLayeredImage to build smaller Docker images in a deterministic and layered approach. If I can build it on one computer with the proper configuration, I can build the same artifact on another one as long as Nix supports the architecture, which in my experience has been very reliable.

That coherence is one of the things I value most about NixOS. The same underlying model helps me with my laptop, my shell, my project dependencies, my CI pipeline and my deployment artifact. It is one way of thinking about software instead of a loose collection of unrelated tools and habits.

So when I say I love NixOS, what I really mean is that I love what it represents. I love a system that is declarative, reproducible, reversible and stable. I love being able to experiment without fear and upgrade without drama. I love that it helps me focus on building and experimenting with fast-moving tools, including LLM coding agents, without worrying about messing up my system in the process.

I love NixOS because it is the most complete everyday expression of what I think software systems should be.