The AI Assistant I Didn't Think I Needed

At first, I was very skeptical. I came across OpenClaw (then called ClawdBot, and renamed twice since) on Twitter, but to me it felt like yet another internet fad. Why would I need to ask questions via Telegram when I could just open ChatGPT? What's the leap forward here?

I tried to install it. Failed. Tried again. Failed again. "This is ridiculous," I thought, and moved on.

Then the FOMO hit. People I actually know and trust started talking about it. Okay, fine. One more honest attempt.

This time it worked. I named it Jarvis (I know, fucking cliché) and started poking around.

The aha moment

I was tracking my food, my fasts, my workouts—but across different apps. I mentioned this to Jarvis. "Could you build me a food tracker?"

And then... it just did.

Without me writing a single line of code, just by chatting back and forth, Jarvis built a system where I could send a voice note—"I had a cheese sandwich with ham and pesto, some peas on the side"—and it would automatically parse the food, estimate the macros, log everything, and show me my progress for the day.

Then I asked for a fasting tracker. It built that too. Then a data pipeline for my health metrics in Apple Health. Then calendar integration. All through conversation.

That was the aha moment. I didn't need to learn an API or write scripts. I just described what I wanted, and the AI built it.

And as a bonus: all the data now lives in one central place—JSON files on my server. It's private. And the data can actually talk to each other in a way it never could when scattered across different apps.

The holy shit moment

A few days in, I was looking at my Apple Health data—six years of daily metrics, workouts, sleep patterns. Jarvis asked: "Do you want me to analyse your health data from the last few years and see what I find?"

I exported everything. 1,481 daily files, 821 workouts. Jarvis processed it all and created a comprehensive report—trends I'd never noticed, patterns buried in years of data that Apple Health just doesn't surface.

It's actually funny

I didn't expect this, but Jarvis has a sense of humour.

One morning I broke a 17-hour fast with a Ferrero Rocher. This was Jarvis' reply.

What changed my mind

An agent that lives on your server, accumulates context over days of conversation, and stores everything in plain files you own? That's different.

The context is what matters. It's not just that Jarvis remembers what we talked about yesterday—it's that this context, accumulated and searchable, turns a generic AI into my AI. One that knows my patterns, my preferences, my goals, because we've been working together.

On privacy

I should mention: everything runs on my own server. My conversations, my memory files, my health data—none of it touches external services unless I explicitly allow it. Even the AI's memory system uses local embeddings now, so the semantic search over my notes happens entirely on my machine.

This was important to me. I wanted an AI assistant, not a surveillance product. OpenClaw let me have both.

What's next? I just gave Jarvis access to a trading account. But that's a story for another post.