Why I built ClawTutor —
and why I'm open-sourcing it

It started with a simple, daily argument. My kids wanted screen time; I wanted them to study first. After one too many negotiations I decided to outsource the whole thing to an AI tutor running on a Raspberry Pi.

👨‍💻
Oliver Gruhn-van Dorp
Creator of ClawTutor

The problem was embarrassingly simple

Every afternoon, same scene: kids arrive home, ask for screen time, I say "do some studying first", they groan, we negotiate, someone gets frustrated. Multiply that by two kids and five school days a week and it adds up to a lot of unnecessary friction.

I'm a software person. I build things. So I started thinking: what if I didn't have to be in the loop at all?

"What if the negotiation just... didn't happen anymore? What if screen time was simply something you earned — automatically — by doing the work?"

Enter OpenClaw

I'd been following OpenClaw — an AI agent gateway for Raspberry Pi by Peter Steinberger. It lets you wire up Claude-powered agents to messaging channels like WhatsApp or Telegram with surprisingly little friction. No custom servers, no complex infrastructure — just a Pi, a config file, and a workspace directory per agent.

The idea crystallised quickly: give each kid a WhatsApp contact — their personal AI tutor — that quizzes them on school topics and rewards correct answers with screen time credits. When credits run out, the TV boxes and gaming consoles go offline via Pi-hole DNS blocking. No drama. No negotiation. The system handles it.

What I actually built

The current setup running in my home:

It's been running for a few weeks now. The daily argument is gone. The kids message their tutor themselves — because they want the screen time. That's the bit I'm most proud of: the motivation is intrinsic now. The system just enforces the deal.

Why open source?

A few reasons.

First, I didn't build any of the hard parts. OpenClaw, Pi-hole, Anthropic Claude, Raspberry Pi OS — I'm standing on the shoulders of a lot of remarkable open source work. It felt right to give back in kind.

Second, I suspect I'm not the only parent who'd want this. The setup is fairly technical — you need to be comfortable with a Linux command line, DNS config, and some JSON — but I've tried to document it well enough that anyone with the relevant background can replicate it. The setup guide and a CLI wizard are already there.

Third, I want to keep building this. Geography tutors, science, sibling challenges, Home Assistant power control — there's a lot of runway. Doing it in the open means others can contribute ideas, catch mistakes, and maybe even submit a pull request.

What's next

The immediate roadmap is mostly about depth, not breadth:

I'll write about each of these here as they come together. If you're building something similar, or have ideas, open an issue on GitHub — I read everything.


— Oliver, Feb 2026

← All posts Gaming consoles under control →