lanowl

Using it

Asking the owl

Ask lanowl anything about your network, in your own words: on Telegram (any message that is not a command) or in the dashboard's Ask the owl tab. The owl answers from the network's data and lanowl's own records, and runs the diagnostics the question needs.

Why did the internet drop on Friday?

🦉 Hoo. The fibre failed on Friday at 19:41 and the router moved the traffic to LTE within a minute; the fibre was back at 19:47. Nothing answered from the internet for the 45 seconds LTE took to come up; after that every device kept working.

Good questions are the ones you would ask someone who had watched the network all week:

  • "Why did the internet drop last night?"
  • "How has the boiler been this week?"
  • "Who joined the guest Wi-Fi today?"
  • "Is the VPS being attacked?"
  • "What did you send me on Sunday, and why no digest this morning?"
  • "Show me the exact log of the home server around 03:00."
  • "Is the attic access point linked at 1 Gbps?"

How it answers

It looks first, then answers: a device's history, the internet's evidence, the router's log, a machine's log, MQTT values, and everything lanowl keeps (every message it sent you and why a digest was held back, the update check, the security review, the backups, earlier conversations). It reads a device itself too, with the login lanowl has for it: the speed a port linked at, the device's own log, its Wi-Fi clients (what each kind offers), and writes a command of its own for the rest: one that only reads runs at once, anything else waits for your button (what it can run). For more it runs diagnostics from a fixed catalog: mtr, DNS, TLS, a scan, a packet capture, a check from the router, from another LAN host or from your VPS. Asking is the approval: they run at once, and each is listed under the answer. Anything that would change something is only proposed, with a button (Actions and approvals).

An answer takes a minute or two. The model does one thing at a time: a question asked while the hourly audit is thinking waits for it, and says so.

You watch it work, on the dashboard and on Telegram: its thinking, each check as it runs, and the answer as it is written. Stop stops it at once. The checks it made stay under the answer, each with the owl's 🪶, in a quote ("6 checks · 32s") that shows the first few: tap it for the rest.

On the dashboard Ask again retries, and Copy copies the answer. Conversations are kept across restarts. The Ask tab opens on a new chat when the last conversation is more than a day old, with up to four questions made from what the dashboard already shows (a device that is down, the last internet drop, an open security finding, a new device, updates waiting); tap one to use it.

On Telegram the work is one silent message, kept up to date, with Stop under it. The answer arrives as a message of its own, so the phone buzzes once, and the working message is deleted. A chat is a conversation too: a question carries the last few exchanges, so "and yesterday?" works. /new starts over; six hours of silence does too. Reply to any message in the chat (swipe on it) and the owl gets that message with your question, and the part you selected, if any: "why?" under an alert is about that alert, and lanowl adds whether its incident is still open and what became of the proposals on it.

Memory

The owl keeps notes that outlive a conversation: what only you know.

  • "Remember that the NAS sleeps from 01:00 to 07:00."
  • "The office's public address is 198.51.100.7."
  • "Forget the note about the old printer."

Notes are written only in your own conversations, when you say so or when it learns something worth keeping. Every change is printed under the answer by lanowl itself, so a note cannot be added unseen. They are read by every model run: questions, the audit, the weekly review, the log checks. A note is context, never a rule: it can change how the owl reads a log, never what is watched or what pages.

You can also edit them directly: /memory, /remember, /forget on Telegram, or the Memory list on the Ask tab.

Its track record

How far can the owl be trusted? Each cause it gives for a problem is checked once the problem is over: the owl is asked again, with hindsight and the evidence, whether the cause it named is what really happened: right, partly, wrong, or can't tell. Your own verdict, one tap on the dashboard, beats its grade. The totals are on the Ask tab and in the weekly review (scorecard).

Written fixes

For each finding of the morning security review, the owl writes the exact fix for that machine: what to type, where, in which order so you cannot lock yourself out, how to check it worked, and how to undo it. Nothing runs: you read it on the Security tab (Fix), copy it, and apply it yourself, or not (fixes).

Over MQTT

A question published on lanowl/ask is answered on lanowl/answer, for an automation that wants the owl's view (MQTT topics).

Edit this page on GitHub.