How I use AI in my personal life

Hello! I’ve iterated a lot on the way I use AI in my personal life, and I’d like to share it here. I have used various browser AI chatbots since their rise in 2022, and in the last year my personal systems and agents have really evolved. Overall my approach is simple: keep my personal data in ordinary local files, teach an agent how to work with those files, and avoid depending on any one AI product.

First, a definition: by “agent,” I mean an AI coding tool that can operate my computer, read approved files, run programs, and use accounts or API keys I give it.

I’ve incrementally expanded capabilities such that now my agents can help me track my finances, analyze my gym workouts and help me get stronger, review my medical test results and research supplements, plan travel for me, and much more.

These tasks all support my broader life goals. I want to build a strong body, see the world, read good books, spend time with my friends, start a family, etc. There are little drudgery tasks on the way to achieving these that I either spend my time on or skip (to my detriment). Reviewing my finances is a great example: I would often miss reviewing my credit card statements for months and then have a huge backlog. Then I made it easy with AI, and now I do it reliably every month and I’m on top of it. Yay!

Let’s start with my hardware: I jumped into the OpenClaw craze of spring 2026, and purchased a Mac mini to run it. It quickly became a useful general-purpose home server. I set it up with its own Apple ID and Gmail account, so it doesn’t touch my personal accounts. I’ve seen lots of criticism about purchasing a Mac mini for OpenClaw “just to make API calls”; why not just run some little server in AWS that runs everything, and can scale as needed? A few reasons:

1. Aesthetics! I like having a pretty little silver box sitting in my room, humming along, where all my data lives. I can see it, I can unplug it, it looks nice. Alex Hormozi has a fun line where someone said his swimming pool was a waste because he never used it. He retorted: “I use it several times a day, every time I walk by and it looks nice, I’m using it.” Aesthetics are real!
2. Apple ecosystem integration. I can share files via iCloud Drive from my phone, and my agent can receive them. I use Apple Reminders a bunch, and I can give my agent access to that, instead of setting up some exotic data pipeline or switching to Todoist. It can send and receive iMessages, which lets me make a group chat with it and my fiancée, and she can ask it questions about wedding planning or our upcoming travel plans.
3. Local compute. I got a cheap bathroom scale that has some awful app, and instead of needing to use the app, I had my agent scan for its local Bluetooth signal, reverse engineer the protocol, and connect to it. Now whenever I step on the scale, the Mac mini automatically logs the reading to a local database and presents it on a little Cloudflare Tunnel web server dashboard I set up. Try doing that on AWS!
4. No monthly server bill. If I were paying monthly for AWS, there would always be a small voice in the back of my head telling me to cancel it or reduce the cost, no matter how trivial it was. The incentive is to stop using it. But if I’ve already paid for the hardware up front, the incentive is to maximize my use of it, to get my money’s worth! I want to enjoy and lean into the latest and greatest AI tooling, and I use my own psychology to make that happen.

The system became useful gradually. My first challenge was figuring out how to give the agent enough information about my life, without locking that information inside another service. I landed on Obsidian.

Obsidian Vault

Obsidian is a note-taking app that sits on top of Markdown files. It has a nice editor interface and a few bells and whistles, like a plugin ecosystem.

I’ve used a lot of different note-taking and tools-for-thought software over the years¹, and Obsidian strikes a great balance: it’s simple and mostly just gets out of the way. I used to have documents scattered across Apple Notes, Notion, Google Drive, and my local filesystem, and they’ve gradually converged to my Obsidian Vault. Here are a few examples of what I store there:

• Every medical document or test result I’ve received, along with AI-generated highlights
• Monthly credit card spending CSVs, along with associated summaries and categories, and higher-level financial documents like tax returns and brokerage statements
• Thousands of bookmarked tweets or blog posts, downloaded as Markdown with their images and videos²
• Misc journal entries, planning docs, and essays in various states of completeness (including this one!)
• Paperwork and contracts: leases, DMV and car maintenance invoices, etc

This documentation helps my agent understand my life and help with all sorts of things. I did a DNA test, gave my agent the sequence (a 10 MB text file), and it curated a list of daily vitamin supplements using my DNA and all past health issues and tests. I started taking those daily vitamins, and several chronic health issues improved. Yay!

Once I started regularly using Obsidian, it got easier and easier to compound the benefits. Another example: I can send my agent a website link and it’ll scrape the webpage and add it to my bookmarks, formatted the way I prefer. I can then have it use qmd³ to do research on a topic across my vault, and write a research report. It’s all saved locally and remains available even if the original is later removed or paywalled.

I can view all this data on my phone or laptop with the Obsidian app, so if I ask my agent to write a document, it’s automatically synced everywhere for easy access. For example, I could ask it to review my travel plans, find the best flights, and suggest activities at various locations. It can check my airline-mile balances, draw on past trips, and write a document directly back to my vault.

What’s really great here is that data storage, computing power, and AI context windows are all massive compared to what you need when you’re working with personal data. If you’re building a scalable AI web application, there are annoying concerns about databases and compute availability and such. But with just your personal data, you don’t need to worry about any of that.

All my journal entries combined over the past decade are approx 25k words, which would be annoying for a human to read, but trivial for an AI to analyze. All the links and X posts I’ve bookmarked add up to about 1.5M tokens, maybe 10 prompts if we’re being conservative.⁴ Personal data is just really small scale!

So: Obsidian handles my personal data, but it’s just a bunch of Markdown files. The other half of this recipe is…

My Personal Collection of Tools

I have this collection of scripts and website code that has evolved to become a sort of personal monorepo. I have a bunch of Skills (aka text files and associated scripts) that detail how to work with my saved bookmarks, how my Cloudflare website infrastructure is set up, how to analyze my monthly finances, how to send content to my Kindle, and more. Similar to Obsidian, once I got the ball rolling, I started seeing more and more things I could do with it. A few examples:

1. I use Voicenotes, which has an API. I gave my agent an API key and had it write a little skill document on how to access my voice recordings, and now it can search over all my past content. I will often take long voice notes, either by myself or during a conversation with a friend, and this gives it a huge corpus of context to pull from when I need help with something. Concretely, when I visit a wedding venue with my fiancée, we take a voice memo on the drive home and have a wide-ranging conversation on our impressions and thoughts, which is all automatically saved and easily searchable later⁵.
2. I use the Hevy workout app, which also has an API⁶. Instead of relying on the app’s various dashboards to monitor my gym sessions, I can just give my agent an API key and it can download all my sessions and give me feedback. Even better: since all my Cloudflare infrastructure is also in this repo, it can wire up dashboards that show my progression, what muscle group is most undertrained, and anything else I want to analyze.
3. Even the website you are visiting right now is powered by this agent! This essay is simply a file in a special public directory of my Obsidian vault, which is automatically published via Cloudflare Tunnel to my website. Any changes I make to the file are immediately reflected. If someone reports a bug on the website, I simply ask my agent to fix it: it edits the files, tests it locally, deploys the change, verifies the fix by visiting my actual website with a browser, and commits and pushes to GitHub.
4. Purpose-built websites become very easy to make. I’ve spun up fun little custom websites for events, which can be a glorified flier or something more involved (take in user input for carpooling optimization, for example).

All this infrastructure has naturally evolved over time, and now adding an additional feature to one of my websites is low friction. I’ve had people suggest an upgrade or point out a bug at a party, and I’ve been able to have a fix deployed within a couple of minutes, with very little effort from me. I once even did so by sending my agent a message from my Apple Watch, which I’ll admit is a bit dorky even for me.

Data Privacy & Security

How you handle your data is a personal decision. My files are stored locally, but that does not mean all processing happens locally. When I use a hosted model, selected data may be sent to that provider. I have decided that letting AI access personal information (tax returns, journal entries) is sufficiently valuable for me that I accept the risks. There are still things I don’t do: my agent has its own Apple ID and its own Gmail account, and it can’t access mine. Bank logins never touch my agent. I pay a lot of attention to what libraries and dependencies my infrastructure uses, and I enforce dependency cooldowns as a protective measure against supply chain attacks. This could be a much longer discussion, but I’ll just note that you should make your own decisions here.

OpenClaw vs Others

I’ve mentioned OpenClaw a few times here. It got me started down this rabbit hole, inspired me to get a home server, and I still use it. However, at the time of this writing I’ve found ChatGPT’s Remote feature to be quite good, and mostly better than OpenClaw. Claude has similar Dispatch and Remote Control features. I wasn’t that impressed when I tried them, but I expect they’ll improve. New products will pop up over the coming months, and I’ll keep experimenting.

The important thing is to avoid getting locked into any specific system. Choose tools that let your AI retrieve and combine your data, while keeping that data portable enough to survive any particular AI product. Obsidian is just Markdown files. Skills are just Markdown files and Python scripts. My websites are just a bunch of HTML files. This makes it easy to swap out AI providers depending on who offers the best experience. I imagine in a year or so I’ll just be self-hosting some Qwen model on a Mac Studio, and all my data will stay nice and local and private. Intelligence is commoditizing shockingly fast!

The most useful part is the collection of ordinary files, instructions, and small programs that has accumulated around my life. Each new piece makes the next use case easier to build, while remaining portable to whatever AI product comes next. You don’t need to construct the whole system at once. Start with one body of data and one recurring frustration, give an agent limited access, and see what becomes possible.