What you're actually building

Not a chatbot. A persistent AI organization that lives in a folder on your computer, structured like a company — with you as the owner and your AI as the leadership team.

You
[Your Lead Persona]
Personal Director
Professional Director
Knowledge Officer
Research Chief
Tech Lead
Creative Director

You talk to one persona. It delegates to the rest. Every one of them is a specialist you can also talk to directly.

The Basics

What is this, in one paragraph?

It's a folder on your computer that Claude treats as its own working memory and workspace. Inside it are instruction files that give your AI a name, a personality, a defined role, and a team of specialists underneath it. There are folders for the different areas of your life — personal, work, research, projects — and the AI reads, writes, and organizes them for you. Over time it accumulates real knowledge about you, and it never starts from zero again.

How is this different from just using Claude or ChatGPT?

Three things:

  • It remembers. Normal chats forget everything when you close them. This system writes what matters to files, so every conversation starts with full context on your life.
  • It has hands. It can read and write actual files on your computer, run programs, search your notes, build things. Not just talk.
  • It's organized like a team. Instead of one generic assistant, you get a lead persona who delegates — research goes to your research chief, filing goes to your knowledge officer. Each one has its own defined expertise and voice.
Where do I actually interact with it?

The terminal. This is the part most people miss, so it's worth saying plainly: your assistant does not live in a web browser. It lives in the Terminal app (Mac) or PowerShell (Windows). You open that window, type one short command, and you're talking to your assistant.

It looks intimidating for about ten minutes and then it doesn't. You're not writing code — you're typing sentences in plain English, same as any chat. The setup walks you through creating a one-word shortcut so starting it up is a single word.

The claude.ai chat you use during setup is only for setup. After that, it's the terminal.

Do I need to know how to code?

No. You'll copy and paste a few commands during setup, and Claude tells you exactly what each one does before you run it. After setup, there are no commands at all — you just talk to it.

The system files are written in plain Markdown, which is just text. If you ever want to edit your assistant's personality, you open a text file and change a sentence.

How long does setup take?

About 30 minutes to get running with a lead persona, a couple of team members, and a working folder structure. A fuller build — the whole team, a database, context files for every area of your life — runs two to three hours if you want to do it in one sitting.

You don't have to. The system is designed to deepen over time. Most people start small and add as they go.

Privacy & Security

Where do my files live?

On your computer. In a normal folder you can open, browse, back up, or delete like any other. Nothing is stored on someone else's server, there's no account holding your data, and no app you're locked into.

If you ever want out, you drag the folder to the trash. That's the whole exit process.

Is Anthropic reading my personal files?

When you ask your assistant something, it sends the relevant parts of your files to Claude to process the request — that's how any AI works, and it's the same as pasting text into a chat window. What it doesn't do is upload your folder wholesale or sync it anywhere.

Anthropic's policy is that consumer conversations aren't used to train models unless you opt in. If you're handling genuinely sensitive material — client financials, medical records, legal documents — treat it the way you'd treat any cloud service and check the current terms yourself at anthropic.com/legal/privacy.

Can it delete my stuff or run things without asking?

Claude Code asks permission before it modifies files or runs commands, and shows you exactly what it's about to do. You approve or decline each one. You can loosen those settings later once you trust it, but it starts cautious by default.

Practical advice: keep your assistant folder in iCloud, Dropbox, or Google Drive, or on Time Machine. Then a mistake is an undo, not a loss.

What should I not put in here?

Passwords, API keys, and account credentials in plain text. Not because the system is unsafe, but because anything sitting in a plain text file on your computer is only as protected as your computer is. Use a password manager for those.

Everything else — notes, journals, project plans, client context, goals, half-formed ideas — is exactly what this is built for.

Practical

What does it cost?

A Claude Pro or Max subscription. Claude Code — the tool that runs your assistant in the terminal — requires a paid plan; the free tier can't run it. Pro is the entry point and is enough for most people. Max is for heavy daily use.

There's no separate cost for this system. The starter kit is free, and the folder lives on hardware you already own.

Mac or Windows?

Both work. The only difference is a couple of install commands, and Claude adapts to whichever you tell it you're using — it asks in the very first message.

What can it actually do once it's running?

A sampling of what people use theirs for:

  • Brain-dump a journal entry and have it mined for action items and routed to the right area
  • Keep a live project tracker current so you never have to maintain one
  • Run real research and file a written brief you can come back to
  • Draft in your voice — emails, posts, presentations — because it's read enough of your writing to know it
  • Build things: scripts, small apps, automations
  • Give you a daily brief that actually knows what's on your plate

It grows with what you feed it. The first week is useful. The third month is a different tool entirely.

What if I mess it up?

Almost nothing here is unrecoverable. The system is text files in a folder — if something breaks, you can ask your assistant to fix it, and it can read and repair its own configuration.

If setup itself goes sideways, paste the error into the chat. That's what the setup prompt is built for — it troubleshoots with you rather than handing you a checklist and wishing you luck.

Can I change my assistant's name or team later?

Any time. Names, personalities, roles, the entire org structure — all of it is editable, and the easiest way to edit it is to just tell your assistant what you want changed. Adding a new specialist to your team is a sentence, not a project.

Ready?

Five steps, about thirty minutes.

Start Setup