It started with FOMO.
Everyone around me was talking about using AI tools. I felt like if I didn't figure this out fast, I'd be left behind. So I thought — let me just try building something.
That "something" turned out to be 6 automated bots running 24/7 on a Mac Mini M1 server.
Here's what actually happened along the way.
The first real mistake
I was building a crypto alert bot for Upbit. Lost a full day's worth of data because I didn't handle API errors properly. The external API changed its response format without notice — my bot silently died.
Didn't notice until the next day.
I've since rewritten that single bot 48 times. v48 is running now.
What "running 24/7" actually means
Spoiler: it doesn't just run by itself.
I discovered Supervisor — a process manager that keeps bots alive with auto-restart. Before this, bots would crash and stay dead silently.
[program:upbit_bot]
command=python3 /Users/m1dongeun/bot_upbit/upbit_v48.py
autostart=true
autorestart=true
stderr_logfile=/var/log/upbit_bot.err.log
Telegram notifications helped too. Now I know about crashes the same day — not the next morning.
The infrastructure stack
- Mac Mini M1 — always-on server
- Supervisor — process manager, all 6 bots
- Tailscale — SSH remote access from anywhere
- Telegram — real-time bot notifications
- Python — everything
What I'm running now
- Upbit pump alert bot (v48)
- Upbit auto-trade bot
- Binance trading bot
- News analysis bot
- YouTube monitoring bot
- Community auto-poster
One unexpected discovery
I built a news analysis bot assuming news would correlate with price movements. So I actually ran the numbers — 1,056 entries, average price change in the hour after news: -0.012%.
I built a whole bot around an assumption that the data didn't support.
Still figuring things out
The latest thing I set up was OpenClaw — an AI agent running on the Mac Mini. Got hit by rate limits on free models, Telegram 409 conflicts from multiple instances running at once, and a launchd vs Supervisor conflict that required going directly to the Mac Mini terminal to fix.
Eventually stabilized it with Anthropic API directly connected and Supervisor managing the process.
Is anyone else doing this solo?
I'm not a professional developer. Everything came from building and breaking things.
If you're running your own automation stack — what's the most annoying failure you've run into?
Top comments (0)