AIBOS
Docs menu

Getting started

Your first hour

A practical walkthrough from intent to shipped work.

4 min read

Your first hour

The sphere is raised, and its health check came back clean. Here's what to actually do with it — a real walk from opening your first session to the nightly caretaker running while you sleep.

Open Claude Code or Codex inside the sphere

Claude Code and Codex are AI coding agents — tools where AI stops being a chat window in a browser tab and starts working directly inside your files: reading them, writing them, running commands. Open one of them with your sphere's folder as the working location. Everything AIBOS does assumes the session starts from inside the sphere: the entry-point file, the rules, the memory, all found relative to where you opened the session. Starting a session from the wrong folder is the single most common way people get confused about where things live.

Meet your lead agent

The first agent you'll actually talk to is your lead agent (the standard calls it the HeadOfSphere) — the one every sphere is born with, created at the same moment as everything else during installation. It introduces itself, roughly, as your co-creator and thinking partner: the first agent you get, and for as long as you want, the only one you strictly need.

Your lead agent works in different modes depending on what you're asking of it — sometimes it's a conversation partner thinking something through with you, sometimes it runs a longer job further on its own, sometimes it handles routine outside correspondence for you. Which mode it uses, and how much it just does versus how much it checks with you first, follows a simple rule: small, reversible actions inside its own job, it just does. Anything with real consequence, it surfaces to you before acting.

Walk the learning track before you start

Before you throw real work at it, run the onboarding command (/teach — Norwegian, meaning "teach"). It's already installed the moment your sphere is born, precisely so your first hour isn't a wall of documentation you're expected to read cold before doing anything. It walks through a fixed order: what a sphere actually is and how ownership nests inside it, how to find your way around, why nothing gets written to disk before its home is decided, which rules are actually enforced rather than just suggested, what add-ons are available and how to add one, and finally what the safety guardrails stop and what running agents actually costs you.

It remembers where you left off. Come back to it in a later session, and it picks up rather than starting over.

Declare where your first piece of work belongs

Before any agent writes a single file into your sphere, it has to answer one question first: has anything like this been touched before? If the answer is no, the work becomes a brand-new thread — a durable line that will follow this piece of work across every session and every agent that touches it from here on. If the answer is yes, the agent decides whether this extends work already underway, opens a genuinely new angle on something already mature, or ties two existing threads together.

This is a real gate, not a suggestion. In careful spheres, the tool will simply refuse to save any work that hasn't declared where it belongs — no file gets written until that question is answered. In practice, this means your very first real request to your lead agent — "look into X," "build a prototype of Y" — gets sorted into a thread before a single file lands on disk, and you watch it happen live, right in the session.

Save what you learned before you close the session

Work in a session happens the way you'd expect — you ask, the agent researches, drafts, builds, and checks in with you at the points that actually matter. When you reach a point worth keeping, run the save command (/lagre — Norwegian for "save"). Think of it as a checkpoint you can hit any time, not only when you're wrapping up for the day. It reviews what just happened for anything worth remembering, writes that to memory, and saves your progress properly, version history and all. You can run it several times across one long session.

One thing worth knowing up front: it's built to actually look for something worth saving, every time, rather than defaulting to "nothing happened here." Even a moment of real frustration in how a session went gets treated as something worth capturing — because knowing what went wrong is often more useful later than knowing what went right.

What happens while you sleep

Every sphere gets exactly one caretaker — a nightly job, running automatically on the schedule the installer set up. It never holds a conversation with you and never makes a decision on your behalf. Its job is narrower and quieter: tidy up loose ends, check the sphere against the standard it's supposed to match, and leave a short report waiting for you. Part of that report is the nightly health check, telling you exactly how far your sphere currently sits from the version of AIBOS you've chosen to follow.

You come back the next morning to a sphere that's been looked after while you weren't watching.