xo stands for:
This is an intimate and kind of weird relationship with an LLM. Don't overanthropomorphize it, and don't think about it too hard.
This is a loose kit to get you started, not a straitjacket.
My operating assumption in writing it is that you have ADHD, you're kind of a mess, you're aware you need better structure and better organization, but you're not really sure what that looks like for you. So xo is going to make some assertions to fill that presumed gap.
When you go, “No, I want to do it my way instead,” do it with my explicit blessing.
Ultimately, the only way this works is if you build it bespoke for yourself, around yourself.
xo starts with a message you paste into Claude Code or Codex. You can try a task you choose before setting up a folder for your instructions and memory.
This is intended for people who are starting from a base where they maybe have used chatbots but have not done agentic work previously.
If you're more advanced than that, then you can probably skip ahead.
This is not a code framework.
This is not an app.
This is not a larger holistic system for running your life.
It will not itself run errands for you or apply to jobs for you or email your mother.
It will, inshallah, provide a logistical framework that helps you do all of those things.
Everything you say or show to your agent goes to the AI company that runs it: that's how cloud LLMs work.
xo will help you find the provider's training setting when you get started. Turning training off does not erase what you've already sent or necessarily shorten how long the provider keeps it.
xo's memory stays on your computer in plain and portable file formats.
You'll need access to Claude Code or Codex in the desktop app; check your existing plan before spending anything to test this.
Start with a task here. Saving your place takes some setup; connecting other apps is optional and can wait.
The apps have usage limits and we might max out a five-hour window while setting up. If we do, that'll be annoying and you'll have to pick up later.
AI makes mistakes and you'll have to correct it sometimes; I feel that these are valuable opportunities to demonstrate continued human relevance but they are, also, very annoying.
This is brand new and not everything is tested yet. Anyone who wants to try it is welcome; tell me where you get stuck.
Pick Claude or ChatGPT, open Code or Codex in its desktop app, and paste the starting message.
This isn't a big decision. You can switch models, apps, or both later without losing ~anything.