Edge AI at Home: Controlling My PC with Raspberry Pi + AI Agent via Telegram

Some people have houseplants on their desk. I have Isolda.

Isolda is a Raspberry Pi 3 that has been living next to my monitor for the past year, collecting tasks like a silent, underpaid assistant. First she ran our internal news digest. Then she got a Claude CLI integration. And now — now she controls my entire desktop PC via Telegram, and the use cases have gotten... unexpectedly wholesome. And also not wholesome at all.

Let me explain. 🙂

The Original Problem (A Boring But Real One)

I travel. Not exotic travel — mostly Wrocław to somewhere, back, repeat. But I often need to make a quick fix to something running on my home PC: restart a service, check a log, open the project in the IDE, run a script.

Carrying a laptop everywhere for "just in case I need to SSH into my home machine" felt excessive. My phone is always with me. My Telegram is always open. So the question became: could Isolda sit at home, stay always on, and act as an intermediary — letting me control my PC through a Telegram chat, with an AI agent doing the actual interpretation of my commands?

The answer, as it turns out, is yes. And it took a weekend.

How It Actually Works

Here's the technical picture, kept short because the funny part comes next.

The setup:

• Raspberry Pi 3 is permanently powered on, connected to the home network
• My desktop PC supports Wake-on-LAN (WoL) — Isolda can send a magic packet to power it on remotely
• Once the PC is on, Isolda connects via SSH to execute commands
• An AI agent runs on Isolda, listening to a private Telegram bot
• I send a message like "turn on the PC and open VS Code" — the agent interprets it, wakes the PC, connects via SSH, and executes the sequence

The agent uses Claude (of course) to interpret natural language commands into actual shell actions. It knows which commands are safe, which require confirmation, and which should absolutely never run without me typing "yes I'm sure" twice.

For screen interaction, it can trigger scripts that take a screenshot and send it back to Telegram. So I can literally see what's on my monitor from anywhere.

The Bonus Feature I Did Not Expect to Need (But Desperately Did)

Here's the thing about working from home when you have a family: the computer is a shared resource.

My husband is wonderful. He is also deeply committed to his gaming sessions on weekends. And to certain YouTube rabbit holes. And on those days when I have a client call in 20 minutes and he's mid-session in something that cannot possibly be paused right now — I used to have to physically walk over, make sad eyes, and negotiate.

Not anymore.

Now I send a message to Isolda: "shut down the PC"

And it shuts down. 😇

No confrontation. No negotiation. Just gentle, remote, inevitable justice.

The Screenshot Feature Has... Multiple Applications 📸

I built the screenshot capability because I wanted to check on running processes remotely. But once it existed, I discovered it has family applications I had not fully considered.

Application 1: Checking if a long-running script is actually still running, or if the PC crashed.

Application 2: Checking if the person currently "watching something" on the PC is watching something appropriate for the hour and audience in question. 👀

Application 3: And look — Isolda also has access to the webcam. So if you really want to know who is at your computer at 11pm... you can know. 😂

I want to be clear that I have used this power responsibly. Mostly. The webcam shot is mostly a deterrent — like a "I could know" energy that keeps everyone honest.

What Surprised Me About This Setup

A few things I genuinely did not expect:

It's faster than I thought. From sending the Telegram message to the PC being up and a script running is about 45–60 seconds. WoL is quick, SSH handshake is quick, and the agent doesn't overthink simple commands.

The AI layer actually matters. Without the agent, I'd have to remember exact commands, paths, service names. With it, I can say "restart the thing that processes the RSS feeds" and it figures out what I mean. That's not a trivial quality-of-life improvement.

Isolda is extremely patient. She has never once complained about receiving a Telegram message at 2am asking her to check on a process. Unlike some other household members I could name.

The paranoia factor is low. Everything runs on my local network or over SSH. The Telegram bot is private, authenticated, accessible only from my account. This is not a "random people can control your PC" setup. This is a "I can control my PC from anywhere" setup.

Is This a Product? Could It Be?

Honestly, yes — a version of this makes sense for small teams and home labs. The core is simple:

• Always-on edge device (RPi, NUC, old laptop)
• AI agent for natural language command interpretation
• Secure tunnel or SSH for local machine access
• Telegram (or Slack, or Discord) as the interface

For businesses, the use case is slightly different — think "DevOps command center in your pocket" rather than "shut down husband's gaming session." But the architecture is the same.

We've built more complex versions of this for clients — AI agents that manage infrastructure, run diagnostics, trigger deployments via chat. It scales. And it starts, as many good things do, with a Raspberry Pi that someone refuses to throw away.

Isolda's Current Job Description

• Monitor processes: ✅
• Wake PC on demand: ✅
• Shut PC down remotely: ✅ (controversial but effective)
• Take screenshots: ✅
• Take webcam photos: ✅ (rarely used, maximum deterrent effect)
• Silently judge family media consumption habits: ✅ (passive)

She earns her electricity bill. Barely. But she earns it.

If you want to build something similar — either for yourself or as part of a business automation setup — reach out. We build AI agents that actually do things, not just chat.

womanintech #raspberrypi3 #coding #creepoteka #programming