Build your own AI Operating System (AI OS)

An AI OS is a set of plain text files that describes your business in enough detail for an agent to work the way you would.

You point an AI agent, like Claude Code or Codex, at that folder. It reads your files first, then gets to work.

That's why it works. Instead of guessing, your agent starts each task already briefed.

ChatGPT is one app, and every new chat starts knowing nothing about you unless you paste it in.

An AI OS isn't an app. It's a layer of files you own, sitting underneath every AI tool you use.

Every agent you use reads the same setup, which makes changing agents painless.

ChatGPT can use your AI OS too, once you connect the AI OS MCP server.

Every session starts with an empty context window. Your agent brings what it learned in training, but nothing about you.

So you keep re-supplying the same details, and correcting the same mistakes the next day.

An AI OS fixes the cause. Your business lives in context files your agent opens as each session begins, and every correction you give goes into a memory file so it holds.

The Memory guide shows how to set that part up.

Your first version is ready in about 15 minutes.

You install the folder with one command and open it in your agent. Then you run /onboard, which studies your existing writing and asks you 12 questions.

When it's done, your context files are filled in, and a fresh session already knows your business and your offer.

From there, your AI OS grows as you use it, one skill and one connection at a time.

Yes. Every file in an AI OS is plain English you can open and read.

You describe what you want, and your agent handles the technical parts: creating files, connecting tools and running commands.

Your job is to answer questions about your business and check what it writes.

If you hit a term you don't know, the AI glossary explains it in plain words.

Yes, for the jobs you want hands-off. Keep your AI OS workspace folder in a private GitHub repo, and cloud tools can work from it.

Claude Code routines clone your repo and run on a schedule, from an API call, or when something happens on GitHub, like a new pull request. They keep working when your laptop is closed.

Claude Code GitHub Actions runs inside your repo's own workflows, on GitHub events or a cron schedule. The Codex GitHub Action runs Codex from GitHub events too.

Codex cloud runs tasks against a connected repo in the background. And you can run agents like Claude Code, Codex, OpenClaw or Hermes on a VPS of your own, so they're always on.

Step 7 of How to build an AI OS covers scheduling.

Any agent that opens a folder on your computer works. That includes Claude Code, Codex, Cursor, Antigravity and Gemini CLI.

They all follow one rules file. CLAUDE.md holds the rules, and the other 2 files simply point to it.

Claude Cowork, ChatGPT, Grok Bot and other cloud agents can't open the folder directly. They connect through the AI OS MCP server instead.

Start with your context layer.

It's the part every other layer depends on, because it tells your agent who it's working for.

Once your agent knows that, drafts sound like you, skills get sharper and automations stop guessing.

The Context guide shows what goes into each file.

With AI OS Pro, you get the AI OS I run my business on, already built. That includes my workspace, a written guide and video walkthrough to set it up, and the skills and plugins built for it.

AI OS Max includes all of AI OS Pro. On top of that, you get live sessions with me where we go through your own AI OS together.

Both come with the AI OS Command Center, and members of both can earn the AI OS Certificate.

You can compare the two on the AI OS Pro and Max page.

Yes. It stays in places you control, like your computer, your private GitHub repo or your own server. Nothing in it is sent to us.

Your agent works with those files like any other folder you open.

Keys for your connected tools stay in a settings file on your machine, which git ignores. They never go in a chat or a context file.

If you keep your AI OS in git, keep the repository private, because your context holds your business.