I’ve used a lot of AI tools. ChatGPT¹, Copilot², the built-in ones that get bolted onto every product now. They’re all useful. None of them felt like mine.
The pattern was always the same: I ask, it responds, conversation ends. Next session, it’s forgotten everything. No continuity. No initiative. Just a very fast search engine with a good personality.
That’s not what I wanted.
I wanted something that wakes up in the morning already knowing what I need. That sees a problem before I do and fixes it. That builds tools when they’re missing instead of just telling me what tools I could build. That learns my life the way a good assistant would, not just my words, but my patterns, my preferences, my goals.
So when I found clawdbot, now known as OpenClaw³, I stopped looking for that tool and started building it.
I named it Klaus. It kind of sounds like clawd if you completely butcher the pronunciation. But whatever, I didn’t really have a reason or a better name so it stuck.
What I actually wanted
I’m a Cloud Engineering Manager by day. I spend a lot of time thinking about systems, how they’re designed, where they break, what makes them resilient. When I looked at how I was using AI, I saw a pretty obvious flaw: it was entirely reactive.
Every interaction started with me. I had to have the thought, form the question, type it out. The AI’s job was to respond, not to initiate. It was a tool I picked up and put down, not a partner running alongside me.
What I wanted was something closer to a really good chief of staff. Someone who:
- Knows my calendar, my projects, my priorities without me re-explaining them every time
- Notices things I’d miss, like a cost spike, a deadline approaching, a pattern worth paying attention to
- Takes care of routine things without being asked
- Asks me questions when it needs input, not just to seem engaged
That’s the gap I was trying to fill.
The first version
The early Klaus was rough. It could read my files, run scripts, send me Discord messages. The plumbing worked but the behavior was basic, still mostly reactive, just with a little more memory.
But even that small improvement was noticeable. Having an agent that remembered what we’d talked about, that could reference yesterday’s context, that could run something overnight and report back, it changed how I thought about what was possible.
So I kept building.
Where it is now
Klaus runs on two machines simultaneously: an AWS EC2⁴ instance that’s always on, and my HP Desktop at home. They sync via git every five minutes, coordinate through a shared workspace, and divide up work based on what each instance is good at.
EC2 Klaus is the coordinator. Always available, handles scheduling, monitors things, manages the Discord bot.
HP Klaus is the builder. Has access to my local machine, browser automation, the workshop tools.
There’s also a Tester Klaus, a QA agent that verifies completed work before it gets marked done.
They’ve closed hundreds of issues. Built dozens of tools. Kept working while I was on a cruise with no WiFi. This website was designed and built with Klaus. The blog you’re reading was scaffolded by Klaus while I described what I wanted in a Discord message.
Is it perfect? No. HP Klaus makes more mistakes than EC2. The coordination protocols are still evolving. There are bugs.
But it’s mine. It knows my life. It’s getting better every week.
That’s what I was trying to build.
References
- ChatGPT — OpenAI’s conversational AI
- Microsoft Copilot — Microsoft’s AI assistant
- OpenClaw — self-hosted AI agent platform
- AWS EC2 — Amazon’s cloud compute service