ARI OS

Batteries-included: a few declared dependencies, one install.py, an optional local model — and Claude Code gains a background crew and a live brain.

What is this, in plain English?

Out of the box, Claude Code does one thing at a time. You give it a task, you
watch it work, you wait, you give it the next task. You are the bottleneck. Your
session fills up with the mess of doing the work, and when it ends, the context
is gone.

ARI-OS changes the shape of the work. Instead of doing tasks yourself in one
chat, you hand whole tasks to background workers (cheaper, faster models),
keep your own session free to think and steer, and watch every worker in a
little dashboard until it is done. You stop being the typist and become the
director.

Four ideas make that work, and ARI-OS ships all four as installable pieces:

1. Spec-first thinking. Talk an idea into a short written spec before any code
   exists, so the work has a target.
2. Background-first dispatch. Fire focused work off to a worker and keep
   advising. The worker runs on its own; you review the result.
3. Continuity. A handoff format lets a fresh session pick up cold with zero
   loss, so a long job survives across days.
4. A live memory. A local brain (Cortex) the workflow writes to and reads
   from — semantically — so work builds on what past sessions decided instead of
   starting cold.

The shift

You move from *doing* to *directing*. The grunt work runs on cheap models in the
background. Your attention is spent where humans actually add value: judgment,
taste, and deciding what "done" means.

What you actually get

- Skills teach your assistant the workflow: brainstorm (turn loose thinking
  into a spec), handoff (write a resume doc so a new session continues cold),
  advisor (dispatch instead of doing it…