Introduction

You're already directing an AI agent instead of just chatting with one — in Claude Code, in a Claude session, in Codex. On a small project it works great. Then the project grows: you want to turn the experiment into real work, and suddenly you're the one holding it together — folders, files, storage, memory. Maybe you've installed skills, maybe a memory plugin. The agents still make a mess: one writes plan_v2.md instead of opening plan.md and continuing it, and it doesn't pick up the thread from this session, or the one before. To keep the context under control, you keep starting fresh sessions and saving things by hand. This page is where that changes.

What AIBOS actually is

AIBOS is a way to run a company where AI agents do most of the actual work, and you direct them — you decide what matters and step in when a judgment call needs a human. Think of it like an operating system rather than an app you open from your dock: you download a specific version, run one script, and it builds a working structure on your machine — your own agents and your own history, under your own rules — that you then keep building on. When a newer version comes out later, you decide if and when to pull it in. Nobody pushes an update onto you.

The memory problem, and it's already yours

You already know this problem, even without a name for it. A new ChatGPT conversation starts from nothing — you re-explain the background, or you don't, and it guesses wrong. Once a project has grown past its first few sessions in a tool like Claude Code, you've felt the other half of the same problem: work piles up in loose files with no single place where a decision actually lives, so nobody — including you, a week later — can say for certain why something was built the way it was.

AIBOS treats that as a problem to solve structurally, not a fact of life to work around. Every session ends by asking a real question — did anything happen here worth remembering? — and writes the answer into a specific file made for that kind of information, not a chat log nobody will scroll back through. The next session starts already knowing what the last one decided. You stop re-explaining yourself.

Your sphere, in one sentence

Everything you build with AIBOS lives inside a sphere: one folder on your computer that holds your entire AI-run company. When someone says they "raised a sphere," they mean they ran the installer once, and that folder came into existence, already structured and ready to work in.

Who this is actually for

AIBOS is written for someone who wants to run a company by directing agents — not for a professional software developer writing code by hand. You need to understand your own business: what has to happen, in what order, and who's accountable when it doesn't. AIBOS handles the underlying plumbing — where files live, how they're named, what gets checked before anything is treated as done — so your judgment is what you spend time on, not file organization.

Where to go next

Three doors from here, depending on what you're trying to do right now:

One honest thing before you go further: the commands you'll actually type — /plasser, /lagre, /tråd, and the rest — are Norwegian today, because that's the language the standard's authors write in. This documentation explains what every one of them does in plain English as you go, and a fully English version of the standard is on the roadmap.