NucleicThe environment
In development

narOS + nash

The whole world
your agents live in.

An operating system built for agents, and the shell it runs them in. They aren't two products that happen to work together — narOS and nash are one environment, designed from the same starting question: what should a machine look like when the thing using it isn't a person?

narOStools · caches · permissions · a screen to drivenashthe shell, all the way downa builda scriptan installyourcockpitone machine · one shell · one record
Scroll
Two names, one system

Designed together,
because they had to be.

You can't bolt a complete record onto an OS after the fact, and a shell can't promise the machine underneath it. So we built both.

narOS

The machine

An operating system whose user is an agent. Everything it needs is already installed and already permitted, and it's the same machine in a thirty-second sandbox as in a week-old VM.

  • Every major language, compiler, and tool, waiting on arrival
  • Caches warm, permissions sorted, sudo that just works
  • No interactive prompt left waiting for a human
nash

The shell

The Nucleic Agent Shell — not a wrapper around the shell, but the shell narOS runs everything with. Nothing to enable, because there's nothing else underneath.

  • Every command, in every script, subshell, and pipeline
  • What crossed the pipe, not just what was typed
  • Secrets masked on the machine, before the record leaves
Turn zero

The first turn is
the actual work.

Every turn an agent spends hunting a package name, waiting on a toolchain, or arguing with permissions is one you paid for and got nothing from. The environment has already answered all of it before the agent arrives — and answered it identically on every machine, so a green run means something instead of meaning the job got lucky about where it landed.

More on narOS →

BEFOREwhich node?apt-get updatepermission?retry…WITH THE ENVIRONMENT$ the actual worksetup is our problem, not your agent's
turns spent on setup → zero
Nothing to switch on

Watched from the
very first command.

Because the shell is part of the machine, every machine is already reporting — there's no watcher to install, no flag to set, and no forgotten box that turns out to be the one that mattered. A thirty-second sandbox and a runner that's been up all week give you the same complete record: what ran, what it carried, and what came of it.

More on nash →

a commanda scriptan install hooknashalready watchingyour cockpithowever it starts
no setup step to forget
Theirs to wreck

Let them do
whatever it takes.

An agent that has to ask before it installs, edits, or breaks something is an agent you're still babysitting. Here it can install anything and change everything: the machine is disposable, nothing it touches is yours, and every move it made is already on the record. Freedom on one side, certainty on the other — which is what makes walking away a reasonable thing to do.

Why it matters

A place worth leaving your agents.

Straight to work

Everything an agent needs is already waiting. No setup turns, no tokens burned on plumbing.

One answer, everywhere

Passing here means passing there. Where a job ran stops being a variable you chase.

Nothing unseen

Every command, in every script and pipeline, from the machine's first second.

Safe to keep

Secrets are masked before the record ever leaves the machine, so it's yours to share.

Keep exploring
Open beta

Give your agents somewhere worth working.

Nucleic is in open beta. All it asks is macOS 27 on Apple Silicon.

No account required · private by design · runs offline