Faux Bunny R1


Twenty-five years ago I was at a Christmas Eve dinner, my fiancée meeting the whole family for the first time, and we were playing dominoes when something about marzipan got stuck in my head. I don’t remember how it started. I just stopped playing mid-game, everybody staring, and went digging through my mom’s cookbooks looking for a recipe. No computer in the house yet, no smartphones. Just cookbooks, and a table full of people waiting on me to take my turn. My brother-in-law, who’s hilarious, named it later: a marzipan moment. Anytime something gets lodged in my head like that and won’t let go, that’s what I call it now.

I was having one over a Rabbit R1 — the AI gadget everybody’s been talking about. Kept pulling it up on Amazon, kept talking myself out of buying it. So I decided to build my own instead and save the money. Only problem is I’m traveling for work right now, not at home, and I don’t have a drawer full of old phones lying around. I thought about waiting until I got back. The itch didn’t care. On the way back to the hotel from work one day, I swung by Walmart and picked up a forty-dollar Motorola.

What I actually wanted was a personal assistant — something in my pocket that wasn’t my phone, that I could just talk to instead of type into, because I hate typing. Something that would get to know my patterns, keep track of my appointments, maybe eventually hook into my email. Still figuring most of that part out. You push a button on the side to talk to it, it talks back, and the screen swipes between cards instead of open apps. Underneath it runs on Hermes, an open-source agent framework from a group called Nous Research.

Getting there wasn’t smooth. The camera was the first fight — I tried to photograph something on my desk and got back a black square. Took forever just to get it talking to the camera at all. Turned out Hermes doesn’t do image recognition on its own, so Claude, who’d been helping me build the thing, took over that job instead.

The backend was worse. It would crash if I so much as went to the bathroom. Claude’s idea was a watchdog, something that notices when the backend dies and restarts it on its own. That took some doing too. Two apps needed to be signed with the same key and weren’t, so I just sideloaded the mismatched one fresh from F-Droid. Quick fix, once we knew what it was.

Tonight was the field test. I’d had it with me in the car all day between meetings — set it on the seat, went in, walked the floor, came back out to tell it something before I forgot it. It said failed to fetch. I was like, well, crud.

I didn’t try to fix it right there. I just went and got a haircut. The hairdresser asked if I had plans for the weekend. Just work, I said, I’m traveling, so I’ve got Saturday and Sunday too. Same, she said, plus a lot of schoolwork — she’s going back to school. Turns out she’s 45, getting a degree in cybersecurity, wants to end up doing penetration testing. I told her I knew exactly what she was going through. I’m 49. Coast Guard, then insurance, and now I’m starting over into network engineering.

Back at the hotel, we went through the settings one at a time. Battery was already unrestricted — checked that first, made sure nothing was putting it into a deep sleep either. Nothing there. It wasn’t until the very last thing we checked, notifications, that we found it. I guess, best I understand it, the phone is always looking to optimize itself — it watches for apps that haven’t been used in a while and shuts them down to save battery. But if an app is set to receive notifications, the phone knows to keep it open for that. Turn notifications on, and it stops killing it.

The other thing bugging me: every reply was taking about twenty seconds no matter what. I figured it had to be the model — I was running something called Luna, which I’d picked myself during setup because it was 80% off right now, dirt cheap. I thought it might just be too big, so I swapped to Haiku for something faster. Same twenty seconds. Model was fine the whole time. The real problem was that it was starting a brand new session from scratch on every single message instead of keeping one open, paying the full startup cost every time I said anything. Once that got fixed, replies dropped to under a second. Hindsight’s 2020. I probably could’ve just stayed on Luna.

This isn’t the first time something like that’s bitten me. Six years ago, Python class, an assignment on lists, I got stuck on a bug I could not find. Called a friend who actually knew Python. He looked at my code and found it. I was about ready to pull out hair I don’t have. That class is basically the reason I dropped it. R, the language I mostly use now, is a lot more forgiving than Python was.

AI catching something like that doesn’t bother me at all. I’m an idiot, and I’m glad there are things smarter than me. Doesn’t make me mad at all.

It feels good having it done. Something me and Claude built together, and there’s a real high in that. The itch is scratched. I’ve got the device now, I can just play with it, keep upgrading pieces of it as I go. Figure I’ll use it for a week and see. Knowing me, it’ll be longer than a week.