One command, and your server reports to Telegram what's wrong with it.

Aromba is an open-source caretaker for servers. It moves in with one command, looks around, and tells you in four lines. It cleans up overnight, and when something breaks it reads the logs and puts the fix on a button — it proposes, you tap.

curl -fsSL https://aro.mba/install | sh

Ubuntu 22.04+ or Debian 12+, root or sudo. Pairs with one Telegram chat; nothing else to sign up for.

• What it touches: caches, logs and temporary files, and only where a chore's header says so. Every chore runs in a cage that allows exactly the paths it names; a write anywhere else fails and is recorded. Anything it cannot undo asks for a PIN first.
• What it never touches: your data. Never deletes your data. Never executes without a tap — the assistant proposes, and the button is what runs it. Never sends file contents: every line that leaves the server passes a redactor, and the relay keeps no message bodies.
• What the AI costs: a fraction of a cent a question, a few dollars in a busy month. It runs on open models — DeepSeek, GLM, Kimi, gpt-oss — through one OpenRouter key, where a closed model costs four to thirty times as much for the same work. It is paid per question, not by plan: the provider bills your key for what was asked, and a daily limit is yours to set. OpenAI's dots, by comparison, come with ChatGPT Pro at $100, $200 or $500 a month, more work on the dearer plans, and run on OpenAI's closed model only (September 2026).
• How to stop it: /panic in the chat freezes it; aromba panic on the server disarms it; aromba uninstall removes it, asking whether to keep the journal.

Docs · Security · Changelog