A bot built for your job — and kept running.

1 — The bot

Written in Python, around your case, not a template with your logo on it. Telegram bots, scraping and monitoring, scheduled jobs, glue between two services that were never meant to talk.

The source code is yours: you get it, and you can host it somewhere else the day you want to. The hosting is a service you choose, not a lock you are stuck in.

2 — The hosting

• — Runs on my infrastructure. No server for you to provision, no process to babysit.
• — Automated monitoring, 24/7, and it restarts itself when the process dies.
• — And when it stops working without dying. The bot reports that it did its job; silence raises an alarm on its own.
• — The alarms reach you too, by email. Same alert, same moment — you are not waiting for me to tell you.
• — Security updates on the system underneath — the machine, not your bot's own libraries. Watching those is a service you can add, below.
• — When a human is needed, I get on it best-effort — a guaranteed window is available, priced separately.

The word doing the work there is automated. Checks run around the clock and the bot picks itself back up on its own — which is what actually fixes most failures, in thirty seconds, at any hour. The second line is the one most hosting leaves out: a process can be alive and doing nothing, and that is the failure nobody notices.

The best-effort part is me. There is one person here, and some weeks that person is teaching. I am not going to promise you a 3am phone call I cannot keep — and since the alarm reaches you at the same time it reaches me, you will see that for yourself rather than read it here.

Code that was safe when it was written does not stay safe. A library your bot depends on can turn out to have a hole a year later, and nothing on your side looks any different — it keeps running exactly as before, and the day it is in the news is the day someone else already knew.

Watching for that is a service you buy, and it is priced separately. I keep the list of what your bot is built on, check it against published vulnerabilities, and tell you when one lands and what it actually means for your bot — most of the time, comfortingly little. Without it nobody is looking, and I would rather write that here than let you assume otherwise. Whether to update is then your call, and the update itself is quoted when you ask for it: I am not going to bill you for work you did not decide to do.

If best-effort is not enough for you, buy a window. A guaranteed response time is a separate, priced service, and the reason it costs something is plain: someone is paid to be awake for it. That is a different promise from mine, so it is a different line on the invoice — not a favour and not a maybe. Most bots never need it. Tell me what breaking costs you per hour and we will both know whether yours does.

In Europe, and that is checkable

Bots run on Hetzner, in German or Finnish data centres. Not on a US cloud. You get a data processing agreement, and Hetzner is named in it as the sub-processor — so you can put my name in your own records without guessing.

Encrypted in transit and at rest. TLS on every channel, and your bot is set up with its data on an encrypted volume — not the whole disk, the volume your data is on, which is the part that matters when a drive is replaced or retired.

On dedicated that volume is yours alone. On shared it is encrypted too, but separation between clients is by process, not by key. I would rather you knew that before choosing than found out afterwards.

If your bot uses AI, it ships compliant

A bot that calls a language model falls under the EU AI Act, and the transparency duties in Article 50 have applied since August 2026. It ships with the disclosure text written, and with the entry ready to drop into your register of AI systems.

If your bot does not use a model, none of that applies — and I will tell you so rather than sell it to you. Most automation is not an AI system, and pretending otherwise is how you end up paying for paperwork you never owed.