After the genome issue, a lot of you wanted the same thing, and it wasn't the genome. It was the folder I dropped it into: my Health OS. "How do I build one?"

My friend Max Barbier didn't wait for instructions. He built his own in a day:

Max Barbier messaging Max Guerois that he built his own Health OS in a day

So here's where it came from, and how to build yours.

It started with one tweet. In April, Andrej Karpathy posted about a simple folder of text files an AI can read[1]. It clicked: this was how I'd finally make sense of my scattered health data.

Bryan Johnson built his own off the same post[2]. When he and a guy in Paris build the same thing off one tweet, that's where this is going.

By the end, you'll have your own.

What a Health OS is

It's simple, almost stupidly so.

A folder on your computer, one plain text file per part of your health: bloodwork, wearables, genome, family history.

And you keep adding to it. Every new lab, every month of wearable data stacks on top of the last. The files aren't a snapshot. They're a record that gets longer and richer over time.

The folder lives on your machine. You open it in your AI tool whenever you want, and it reads across all of it at once, the whole history, every source, and turns it into decisions.

No app, no dashboard, no code.

Your health files go in, the AI reads across all of them, decisions come out
Your files go in, the AI reads across them, decisions come out.

I had years of data. Nothing read it together.

For years it piled up, in two speeds.

The daily stuff: Whoop on my sleep and recovery, Withings on my weight, Strava on every run and workout.

And the bigger tests, more like once a year: bloodwork from Lucis (the company I built), a body composition scan, my genome, my family's health test results.

Good data, all of it. But each source only ever saw its own slice.

My Whoop didn't know my iron stores were low. My bloodwork didn't know I'd trained hard all week. Nothing looked at the whole thing.

That's the gap AI closes.

It reads across plain, messy files from every source and finds what no single one can see on its own. No database, no integrations.

You hand it your text, and it reasons over all of it at once. The data you already have is enough. It just needs one reader that sees all of it.

Then it stopped being a folder and became a system

I started simple. Dumped everything into the folder and started asking questions. It worked right away, ask anything, get an answer across all my data at once. Things no single app could see.

But I wanted to push it further. The folder was already great. It got better the more current it stayed, and the moment it updated itself every day, it started getting more useful every day too.

The friction was keeping it current by hand. Every Sunday I'd sit down to plan my training, and first I had to:

It worked. But it ate time, every single week.

So I automated it.

And here's the part I still can't quite believe: I didn't write a single line of it. I told Claude Code what I wanted, and it built the whole thing, the scripts, the integrations, all of it.

Now it runs on its own:

Then I took it one step further. I had Claude Code wire it into a Telegram bot, Max Health Coach. Now it comes to me:

Max Health Coach messaging me every morning on Telegram with how ready I am and my workout

That's the whole arc:

a folder I query a system that updates itself a coach that texts me every morning

And I never touched the code.

Why you'd build one too

A few months in, here's what it actually buys me.

If you measure anything about your health, this is what turns the numbers into decisions.

How to set it up, no code

You don't need the automation to start, that came later. If you can edit a text file and paste a prompt, you can do this.

1. Pick a tool that reads a folder. Claude (Cowork or Claude Code), OpenAI Codex, any AI tool that lets you open a folder from your computer and ask questions about the files inside. Free tiers are fine.

2. Make the folder. One plain file per topic:

The simplest version looks like this:

One of those files can be this plain:

# bloodwork.md
Date: June 2026
Vitamin D: 38 (a bit low)
Iron stores: 45 (low)
Testosterone: 620
The Health OS Starter template file structure: profile, data, protocol, instructions, and dashboard files
The Health OS Starter template structure.

3. Add your data. Paste in what you have, lab PDFs, your Whoop or Oura, Apple Health, a body composition scan. Start with bloodwork and one wearable, add the rest later.

4. Give it the coach instructions. This file does the work: read everything first, never invent a number, cite sources, flag anything that needs a doctor. Ask for your read and it tells you where you stand, what's flagged, what to do, what to stop, what's missing. It plans and coaches too.

5. Run it weekly. New data lands, drop it in, say "give me my read." The picture gets sharper. Data you collect is a curiosity. Data something acts on every week is a coach.

I packaged the exact scaffold, the coach prompt, and example (fake) data into a free starter template. Copy it, swap in your data, and open it in your AI tool.

Get the Health OS Starter → github.com/gm-max/health-os-starter

The genome was the base layer. This is the thing that finally acts on it.

If you build one, reply and tell me what you're tracking. I read every reply, and I'll help you get yours set up.

More soon,
Max

Personal experimentation and documentation, not medical advice. Talk to a doctor before making health decisions.

References

  1. Andrej Karpathy on a plain-text notes folder an AI can read
  2. Bryan Johnson on building his own health database