I have a bunch of things running 24/7 on a Mac Mini. GPU rental jobs, a Garmin watch face updater, a Fiverr inbox monitor, a funding rate tracker, a few cron jobs.

For a while I ran a Grafana dashboard to keep an eye on them. It looked impressive. I never opened it.

What I actually do is check my phone. So I built the monitoring layer there.

Here's the setup: a lightweight Telegram bot that serves as my entire DevOps interface. Status checks, alerts, and even simple commands — all from the Telegram app I already have open.

Why Not a Proper Dashboard?

Honest answer: dashboards are for teams. If you're a solo dev with a few projects, a fancy web UI creates more overhead than it solves.

Problems I had with Grafana:

• VPN required to reach it from outside my home network
• Needs to stay running (another thing to maintain)
• I never actually opened the browser tab
• It didn't push me information — I had to pull it

Telegram flips this: it pushes alerts to me. I glance at my phone, see what's happening, and move on.

The Architecture

Services (cron jobs, Python scripts, shell scripts)
 ↓
Central alert script: notify.sh
 ↓
Telegram Bot API → my phone
 ↓ (optional)
Command bot → runs queries on server

Two pieces:

1. Outbound alerts — services send me messages when things happen
2. Inbound commands — I can ask the bot questions from my phone

Part 1: Dead Simple Alert Script

Every service on my server can call this:

#!/bin/bash
# notify.sh — send a Telegram message from any script
# Usage: ./notify.sh "Your GPU job finished"

BOT_TOKEN="your_bot_token_here"
CHAT_ID="your_chat_id_here"
MESSAGE="$1"

curl -s -X POST "https://api.telegram.org/bot${BOT_TOKEN}/sendMessage" \
 -d chat_id="${CHAT_ID}" \
 -d text="${MESSAGE}" \
 -d parse_mode="HTML" > /dev/null

That's it. Any script can now send me a message in one line:

./notify.sh "✅ GPU rental job completed — earned $2.40"
./notify.sh "⚠️ Funding rate dropped below threshold on LYN_USDT"
./notify.sh "📬 New Fiverr inquiry from user987"
./notify.sh "❌ Garmin watch face API returned 503"

I spent maybe 20 minutes on this. It replaced a monitoring stack I spent days configuring.

Real Examples from My Setup

GPU rental monitor:

# Runs every 30 min
earnings=$(./check_gpu_earnings.sh)
if [ "$earnings" -gt "0" ]; then
 ./notify.sh "💰 GPU earned: $earnings today"
fi

Funding rate watcher:

# Python script, runs every 15 min via cron
rate = get_funding_rate("LYN_USDT")
if rate < -0.5: # negative = people paying longs
 notify(f"🔥 LYN funding rate: {rate}% — worth checking")

Daily summary (9 AM cron):

#!/bin/bash
msg="📊 Daily Summary — $(date +%Y-%m-%d)

GPU Jobs: $(get_gpu_count) completed
Funding Earned: $(get_funding_total)
Fiverr Inquiries: $(get_fiverr_count)
Watch Face Updates: $(get_garmin_count)

Server uptime: $(uptime -p)"

./notify.sh "$msg"

I wake up, check my phone, and immediately know if anything needs attention. No browser, no VPN, no dashboard.

Part 2: The Command Interface

Outbound alerts are great. But sometimes I want to query the server from my phone.

I wrote a simple Python bot that listens for commands:

import telebot
import subprocess

BOT_TOKEN = "your_token"
ALLOWED_USER = 123456789 # your Telegram user ID

bot = telebot.TeleBot(BOT_TOKEN)

COMMANDS = {
 "/status": "uptime && free -h && df -h /",
 "/gpu": "./check_gpu_status.sh",
 "/funding": "python3 check_funding_rates.py --summary",
 "/services": "ps aux | grep -E '(python|node|ollama)' | grep -v grep",
}

@bot.message_handler(commands=list(COMMANDS.keys()))
def handle_command(message):
 if message.from_user.id != ALLOWED_USER:
 bot.reply_to(message, "Not authorized.")
 return

 cmd_text = message.text.split()[0] # handle /status@botname format
 if cmd_text in COMMANDS:
 result = subprocess.run(
 COMMANDS[cmd_text], 
 shell=True, 
 capture_output=True, 
 text=True,
 timeout=15
 )
 output = result.stdout[:3000] or result.stderr[:3000] or "No output"
 bot.reply_to(message, f"```
{% endraw %}\n{output}\n{% raw %}
```", parse_mode="Markdown")

bot.polling(none_stop=True)

Now from Telegram I can type /status and get:

 11:23:15 up 14 days, 3:41, 1 user
Mem: 16Gi 8.2Gi 7.8Gi
/dev/sda1 245G 82G 163G 34%

Or /funding and get the current rate snapshot.

The key detail: ALLOWED_USER check. Only my Telegram ID can run commands. Everyone else gets "Not authorized." Bot tokens are public in the sense that anyone can message your bot — you need to validate the sender.

Keeping It Running

The command bot needs to stay alive. I use a simple systemd service:

[Unit]
Description=Telegram DevOps Bot
After=network.target

[Service]
ExecStart=/usr/bin/python3 /home/user/telegram-bot/bot.py
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target

systemctl enable telegram-bot && systemctl start telegram-bot — and it survives reboots.

On macOS (my setup) I use a launchd plist, same concept.

What I Actually Get Alerts For

Not everything. Alert fatigue is real.

Alert on:

• ✅ Job completions (GPU task done, funding cycle closed)
• ❌ Errors that need action
• 📬 New customer inquiries
• ⚠️ Thresholds crossed (rate drops, disk usage, memory spikes)
• 📊 Daily summaries (once a day, morning)

Silence:

• Routine successful runs (no news is good news)
• Health checks that pass
• Regular cron completions with no anomalies

The goal is: every message I receive from the bot is something I actually care about. If I'm ignoring 80% of notifications, I'm alerting on the wrong things.

The Full Cost

• Telegram Bot API: free
• curl command: comes with your OS
• Python + telebot library: free
• Running this bot: negligible CPU, ~20MB RAM

My entire monitoring setup costs $0/month and runs on the same Mac Mini as everything else. No SaaS, no cloud logging, no dashboard subscription.

Three Months Later

I send about 15-20 alerts per day. Daily summary at 9 AM, event-driven messages the rest of the day. I check my phone, see green checkmarks and earnings summaries, and know the server is doing its job.

The one time the GPU host went offline, I got a message within 5 minutes. Fixed it from my phone during lunch.

That's the whole point: not more tooling, just the right interface for how I actually work.

Got a monitoring setup you like? Drop it in the comments — always curious what others are running.