Hey Lemmy! I want to share a project I’m currently working on.

A bit about myself: I live in Norway, got a family and kids. I work at a warehouse driving a forklift. At work, I actually managed to implement a couple of cool productivity hacks—I’ll make separate posts about that later.

But I also have this huge interest in all kinds of hardware, servers, and, like many people right now, tweaking AI. AI is everywhere now, and sometimes we just use it to get a quick answer and run along.

Not long ago (or maybe a while back), OpenClaw came out, then Hermes. There’s probably more out there, I honestly don’t have enough time to keep up with everything. I played around with these tools; each has its pros and cons. But one thing is for sure—they can be a real pain to set up and deploy.

And here, drum roll… I created a unique app!

Nah, just kidding. The story is different. Since I ran into so many deployment issues myself, I decided to try making a lightweight, highly-focused tool that shares a similar concept but is built entirely in my own style.

How it was made To build it, I first mapped out a plan in a simple text editor and drew some tables to figure out what should go where and how it should work. I just needed to understand how I envisioned it and what I wanted in the end. I have some basic understanding of code, but definitely not enough to just sit down and build an app from scratch. Back in the day, I used to build websites and did some media stuff, so I knew a few bits and pieces here and there.

To actually code the app, I cobbled together a specific setup. I have absolutely no idea if I did it the “right” way or not. Sometimes I had to manually tweak a few lines of code, digging for answers on the internet. AI is amazing, sure, but I realized it’s crucial to actually understand what you’re doing, what you want, and see the whole picture from all sides.

My setup consisted of Antigravity IDE, Trae IDE, and OpenCode CLI. The models I threw at them were Gemini, Claude, DeepSeek, and Minimax. I also kept a few active chat windows open just to phrase my thoughts properly. I moved strictly block by block, checking every single function, and backing it up the moment it worked.

My main goal was just to make an app for myself. To use it every single day exactly how I want to. Why am I here then? Not for commercial profit, not for hype, and not for scaling. Tomorrow morning I’m heading right back to the warehouse, I have a ton of work waiting for me there. But I just want to touch this tech environment that always runs parallel to my life as a hobby. I wanted to try making something—who knows, maybe someone else will like it too.

This is getting way too long, so let’s get straight to the point: what did I actually make?

What is Hoofify? I built an Android app called Hoofify. The core idea is basically a mini-version of OpenClaw.

Background Server: The app runs a stable server right on my phone. It doesn’t turn off, and it doesn’t need the screen to be on—it just sits quietly in the background and does its job.

Telegram Interface: This server holds an AI API key (in my case, DeepSeek). I hooked up Telegram via a token, and my connection is ready. I control everything through my private Telegram bot.

Google Integration: To make it a true agent, I connected it to Google Mail and Calendar. Our pocket assistant can perform actions based on a schedule (via an internal timer) or whenever you ask it to. It also understands photos and screenshots through vision models.

Isolated Sandboxes: In OpenClaw, I loved the multi-agent idea where everyone handles their own specific task. But since this runs on a smartphone, I can’t just spawn dozens of heavy agents. To keep one agent’s “brain” from getting cluttered and mixing up context, I built a system of sandboxes. When I give it a task (or if the agent realizes it needs to separate its thoughts), it spins up a dedicated sandbox with its own isolated history. It acts as both a memory bank and a separate folder for everything. Essentially, it’s a multi-agent setup, but wrapped inside a single agent.

Self-Development (Dynamic Skills): Another awesome feature that came together is the agent’s self-learning. I can literally tell it what I need in plain text, and it will write a custom JavaScript skill for itself. It knows how to write it and how to use it properly—whether it’s tracking something specific on the internet or just fetching an account balance via an API.

Design & Nostalgia The entire app is a massive wave of nostalgia for Windows 95-98, which is why it has that specific retro design. I thought: hey, if I´m giving a second life to old phones, why not dress them up in that classic style? To back up the nostalgia, I even have a real, working retro PC from that era sitting at home. Not life, but a pure wonder!

Just being honest I’ll be completely honest for the sharp minds and harsh critics out there: I built this strictly for myself to solve my own frustration. This entire post is just a raw stream of my thoughts. If you see similarities to other products or ideas, it doesn’t mean it’s a direct copy. I might have seen or heard something somewhere, but I don’t have the time to know every single project on earth, and I would never pass someone else’s work off as my own.

I tried to make a tool for myself, but maybe someone else will find it useful too.

I’ve only tested this on my two personal phones, so if you try it on other Android devices, expect some unexpected bugs. But I’m online, I’m here in the community, and we can figure out and patch any issues together.

I would love to get your honest feedback, criticism, and ideas on what to build next!

And yeah, I put it on Gumroad for 3+ bucks—not for profit, but simply to filter out the crowd of people who just want to install it and criticize. Or maybe nobody even needs this, I’m no prophet, that’s just how I think. Maybe someone will find it interesting. I’m not placing any bets here.

Website: https://hoofify.app/

GitHub (Docs & Skills): https://github.com/vnimatiq/hoofify

Good vibes to all! =)