How long does it take to build an AI OS?

The first version takes about 15 minutes with /onboard in the AI OS Starter Kit. After that, your AI OS grows one addition at a time, usually one a week.

Do I need to know how to code to build an AI OS?

No. An AI OS is written in plain English, and your agent handles the technical setup.

Which AI agent should I use for my AI OS?

Any agent that reads a folder. The AI OS Starter Kit is built for Claude Code, and the same folder works with Codex, Cursor and Gemini CLI.

What's the difference between context and memory?

Context is small and loads every session: who you are, what you sell, how you write. Memory is big and gets looked up only when a task needs it: your brain, meetings, decisions and daily notes.

Where should I keep my business data?

In one place your agent can query, such as a database, with tables linked by a shared ID like a customer's email address.

Do I need MCP servers?

Not on day 1. Your context and skills work without any. Add a connector, an MCP server or a CLI, when a skill needs to reach a specific tool.

Where do API keys go?

In a local settings file that git ignores, such as .claude/settings.local.json in Claude Code. Never in a chat, and never in a context file.

How do I stop my AI from forgetting what I told it?

Save it to a file your agent reads every session, such as MEMORY.md. A correction that only lives in the chat is gone when the session ends.

How do I make my AI OS write like me?

Give it 5 to 10 things you actually wrote, plus a list of words and phrases you never use. Then have a checker read every draft against both before you see it.

Can my AI OS run while I'm away?

Yes. In Claude Code, /schedule runs routines in the cloud, at most once an hour, and desktop scheduled tasks run on your own computer. Automate a skill only after it works by hand.

What does an AI OS cost?

The AI OS Starter Kit is free. You pay for your AI agent's plan, and for any paid tools you connect.

How to build an AI operating system (AI OS)

Learn how to build an AI Operating System (AI OS) in Claude Code, Codex, or any AI agent, and use it to automate your business and life. You go from an empty folder to a system that knows you and runs every day.

by Navid Moazzez·Updated 2 de out. de 2026·39 min read·

An AI operating system (AI OS) is the layer underneath your AI tools.

It holds your context, your voice, your offers, your customers and your workflows in one structured place that every AI agent you use can read.

Think of how macOS or Windows loads your files, your apps and your preferences the moment you log in. You don't tell it who you are every morning.

An AI OS does the same thing for your AI.

In this guide, you'll learn exactly what goes into an AI OS, and how to build yours from scratch:

What an AI OS is, and what it isn't
Why your business needs one
The 7 layers of an AI OS
How to build yours, step by step
How to measure it, and the mistakes to avoid

key_takeaways.md

TL;DR

Key takeaways

An AI operating system (AI OS) is the layer underneath your AI tools: your context, your voice, your offers, your customers and your workflows in one structured place that every AI agent you use can read.

In practice, it's a folder of plain text files. You open it in an AI agent like Claude Code or Codex, and the agent reads your files at the start of every session, before you type a word.

Every AI OS is built from the same 7 layers: context, data, tools, memory, skills, automations and apps. Each layer needs the one below it, so you build them in order.

Context comes first. Every layer after it works better because your AI already knows you.

Keep the tools you use, build it by hand before you automate it, and update your files every time you catch something that's off.

Let's start with what an AI OS actually is.

What is an AI operating system (AI OS)?

An AI OS is the layer underneath your AI tools. It holds your context, your voice, your offers, your customers and your workflows in one structured place that every AI you use can read.

In practice, it's a folder of plain text files: your AI OS workspace folder. It starts on your computer, and a copy in a private GitHub repo gives you a backup, team access and the option to run certain jobs in the cloud. You open that folder in an AI agent, and the agent reads your files at the start of every session, before you type a word.

Here's what that folder looks like:

~/my-ai-os

Claude Codev2.1.138

> /onboard

🧠CLAUDE.mdthe rules, read first every session

·AGENTS.mdsame file, so Codex and Cursor read the same rules

·GEMINI.mdsame file, so Gemini CLI reads them too

·MEMORY.mdwhat came up while working

▸context/who you are, your company, your voice, your brand, your tools

·about-me.md

·my-company.md

·voice-profile.md

·anti-ai-writing-style.md

·design.md

·tools.md

▸decisions/what was decided and why

·log.md

▸inbox/what's open, and your setup answers

·open-work.md

·onboard.md

▸daily-notes/One note per day

▸outputs/everything your AI OS makes, one folder per project

▸projects/briefs and references, one folder per project

▸references/the frameworks, SOPs and tool notes your AI OS runs on

▸assets/logos, headshots, fonts, colors

▸_archive/where deleted things go

·.mcp.jsonthe tools your AI OS can reach

▸.claude/

▸skills/the 12 skills

▸onboard/

▸morning-brief/

▸meeting-notes/

▸research-assistant/

▸slide-deck/

▸visual-explainer/

▸diagram-generator/

▸ai-os-daily/

▸ai-os-connect/

▸ai-os-audit/

▸ai-os-update/

▸ai-os-grow/

▸agents/

·brand-guardian.mdchecks every draft against your voice

That's how I run my own business: an AI OS with Claude Code as the brain.

And it isn't tied to one agent. Each agent reads a rules file when a session starts, and you point all of them at the same one:

Agent	Reads this rules file
Claude Code	`CLAUDE.md`
Codex	`AGENTS.md`
Cursor	`AGENTS.md` and `CLAUDE.md`
Gemini CLI	`GEMINI.md`

So you set it up once. Then you can switch agents without starting over.

How an AI agent uses your AI OS

Every AI agent works in a loop. It looks at the situation, decides the next step, takes it, and checks the result. Then it goes around again until the job is done.

~/my-ai-os

lookreads your rules file, your context and the task

decidethe AI model picks the next step

actruns a skill, uses a tool or writes a file

lookchecks the result against your AI OS

repeats until the job is done

Your AI OS is what the agent looks at, every time it goes around. The better that folder knows your business, the better every step.

What an AI OS is not

An app you download

It's a folder of plain files that the agent you already use reads

Something you need to code

It's written in plain English, and the agent handles the technical setup

Tied to one AI

Claude Code, Codex, Cursor and Gemini CLI read the same files

Just a starter kit

The AI OS Starter Kit is where you begin. The system is the layers you keep growing

AI OS vs ChatGPT, custom GPTs and automation tools

Here's how an AI OS compares to the tools you probably use today.

Where it lives

AI OS: Your AI OS workspace folder, on your computer and in GitHub
ChatGPT or Claude chat: Inside that one app
Custom GPTs: Inside one custom assistant
Automation tools: Inside each workflow you build

Works in other AI agents

AI OS: Yes
ChatGPT or Claude chat: No
Custom GPTs: No
Automation tools: No

Works out the steps itself

AI OS: Yes
ChatGPT or Claude chat: Yes, inside one chat
Custom GPTs: Yes, inside one chat
Automation tools: Mostly no, it runs the steps you mapped out

	AI OS	ChatGPT or Claude chat	Custom GPTs	Automation tools
Where it lives	Your AI OS workspace folder, on your computer and in GitHub	Inside that one app	Inside one custom assistant	Inside each workflow you build
Works in other AI agents	Yes	No	No	No
Works out the steps itself	Yes	Yes, inside one chat	Yes, inside one chat	Mostly no, it runs the steps you mapped out

So keep the tools you like. An AI OS sits underneath them, and your agent can run them through connectors.

Navid Moazzez

@thenavidm

Most creators and founders are still stuck in chatbot mode. Build your own AI Operating System (AI OS) instead. The ones winning in 2026 and beyond run their entire business through it.