lanowl

Using it

Updates, backups and the security review

Every morning, lanowl checks what is waiting to be updated, the owl reviews what each machine exposes, and the configurations are compared with yesterday's. Once a month the machines are backed up and scanned for known vulnerabilities. Only what is new and matters reaches Telegram, once; everything is on the Security tab and in /security.

Each part works on the devices whose manage list names it (Devices), with its switch on in config.yaml, on every site. The vulnerability scan needs no login: it takes every device whose scan is on.

Updates

updates:
  enabled: true
  at: "06:30"
  scan: {enabled: true, day: 1, at: "04:00", top_ports: 1000, deny_groups: ["security", "iot"]}

At at every day, lanowl reads each machine's waiting updates, read only: apt on Linux, RouterOS against MikroTik's own latest release (and whether the versions in between were security releases), Home Assistant's update list (which covers its add-ons and the firmware it knows), ESXi's version, OpenWrt's and UniFi's release, your profiles' updates. It also notices an insecure service newly open (telnet, ftp).

What pages, once: a reboot pending for more than 3 days, a security update still not installed after 2 days, a security release or a serious vulnerability for something installed, a newly opened telnet or ftp. Everything else waits on the Security tab.

Installing is an action you approve: Update on the machine's row, /upgrade on Telegram, or the owl's proposal. When an update leaves the machine asking for a reboot, the row says "a reboot finishes them" and lanowl proposes that reboot at once, on the dashboard (Approve reboot on the same row) and on Telegram: it is checked first and its alerts are held while it restarts, as any reboot. A machine lanowl cannot reboot says "reboot it yourself".

The monthly vulnerability scan

On scan.day at scan.at (or Deep scan now), the devices of every site are scanned for service versions with nmap's vulners script: the main network's, a remote site's through its tunnel, your server on the internet at its own address. You choose which devices, one by one: at setup each device has a Deep scan choice (off for cameras, smart-home gear and phones, on for the rest, until you change it), and later its settings have a scan switch. Some cheap devices hang under a version scan; switch those off. A device you never chose for follows its group: the groups in deny_groups are left out. lanowl's own machine and an address on none of your sites' networks (a ping target on the internet) are never scanned. In the inventory:

    - ip: 192.168.88.21
      name: Garden camera
      group: cameras
      scan: false                  # never version-scanned, nor port-scanned by the owl

The daily telnet and ftp check asks every site's devices, switched off or not: two connection attempts, no nmap. A version scan matches many CVEs that are not real (Debian and Ubuntu fix flaws without changing the version a service announces), so each match is checked against fixed sources: the exact package installed, asked of OSV.dev (the distributions' own security data), or NVD's record of the CVE. What is still affected goes to the owl with the machine's own facts: does the flaw's condition hold here? Only one that applies, with CVSS 7 or more, pages. What leaves your network for this is package names, versions and CVE ids, never an address or a name.

The security review

exposure:
  enabled: true
  page_from: "2026-11-01"          # watch what it finds for a while before it pages
  outside_scan: {top_ports: 1000, max_rate: 20}
  public_names: []                 # your own host names, looked at from outside

After the update check, lanowl reads, read only, what decides how exposed each machine is: who may log in and how, what listens, the firewall, which management services answer and from where, UPnP, an open resolver, the firmware's age. It never reads a password, a key, a Wi-Fi passphrase or a password hash. It also looks from outside: your public server is scanned from here, and your home's public address from that server, but only if that address is really your router's (behind a provider's shared NAT, nothing inbound can reach you anyway). Each remote site the same way: when your server is the hub of the sites' WireGuard tunnels, it knows the address each tunnel comes from (wg show's endpoints and allowed IPs: public keys only), and that address is scanned from the server only when the site's router holds it. Behind the provider's NAT, or behind a modem or the provider's router, it is said and not scanned: the address may be shared. The Security tab shows it under the review: Seen from the internet, one line per site.

The owl reads it all and names each problem with one of a fixed set of checks (ssh accepts passwords, a management service exposed, a firmware out of support, and so on), so the same problem keeps the same name every day. That is what lets you Accept it once, with a note. It is accepted until you undo it, or until the problem has been gone for three reviews. The review's summary leaves it out; the Security tab names what you accepted once, under the summary.

Pages start from page_from (watch what it finds for a couple of weeks first), and then only for a new critical finding, once.

Fix shows the owl's written fix for that machine: the exact steps, in an order that cannot lock you out, how to check it, and how to undo it. Nothing runs.

What changed

configwatch: {enabled: true}

Every morning after the update check (or Compare now), each machine's configuration is compared with yesterday's: a MikroTik's /export, a Linux machine's effective sshd settings, accounts, sudo rules, keys, crontabs, services, listening ports, firewall and VPN peers. The owl says what each change does and how risky it is, knowing what you ran on the machine in between. A change in a part that decides who gets in, that the owl did not explain, is listed anyway. The first comparison is the baseline. Nothing pages: a high or critical change rides the next digest and stays in To decide until you mark it handled ("It was me").

Slowly changing

drift: {enabled: true, at: "07:30"}

Once a day the owl reads a week of the sweep's own numbers, per device and per day, and names what is slowly getting worse: a device that drops every night at 02:00, a line that flaps a little more each week. Latency alone is never a finding, nor something already over. Only a high one rides a digest; the rest is on Now and in the weekly review.

Logins and logs

hostlog:
  enabled: true

For Linux machines with manage: [logs] (by lanowl's key), lanowl reads the auth log every two minutes: ssh logins, failed passwords, invalid users, sudo, and kernel warnings. A burst of failed logins, a login from an address you do not trust (logs.trusted), or a flood of connections is an alert at once, without waiting for the model; the owl reads the lines no rule explains. lanowl's own logins are recognised by user and address, so they are never mistaken for someone else. The router's log is read the same way.

Every device with manage: [logs] — a MikroTik access point, OpenWrt, UniFi, ESXi, a Linux machine — also has its own log read once an hour (devwatch; on a Linux machine the journal's errors and kernel lines, and its auth lines too when the two-minute read above does not cover it): a login that does not fit and failed passwords again and again are the owl's to tell, with the device's own lines — at once from a critical device or when it reads like someone getting in, else in the next digest (Device kinds).

Each security event stays in To decide until you mark it handled.

Backups

backups:
  enabled: true
  day: 1
  at: "03:00"
  keep: 6
  keep_before_days: 90
  store: {host: "192.168.88.11", path: "/srv/lanowl-backups"}
  self: {daily: 7}

On day day of every month, every machine with manage: [backup] is backed up onto store: a Linux machine lanowl reaches by its key, on a disk of its own (not the datastore of the VMs it backs up). The last keep are kept per machine, and one is taken before every update (kept keep_before_days). How each kind is backed up: Device kinds.

lanowl also backs itself up daily (self.daily kept): its database (history, memory, actions, conversations, dismissals, pauses), its secrets, and the config it runs with.

Not encrypted. The backups hold your network's passwords and keys in the clear. Keep the store's path as private as secrets.yaml.

One message when a monthly run had a failure, or when a machine has had no good backup for two months. Back up everything now on the Security tab, /backups on Telegram.

Setting it up from the dashboard

Backups start switched off, like every optional feature: until then a device's page shows backup with "switched off in config.yaml", and lanowl --check says ○ backup … off: backups.enabled in config.yaml. Three steps:

  1. The store. In Settings → Devices, open the Linux machine with the disk (or add it): kind linux, its login "its own login: lanowl's ssh key", and the user. Run the line the page gives on that machine, as that user.
  2. Settings → Backups: switch it on, store.host that machine's address, store.path a folder on it that only you can read. The save checks the store first: a machine that is not one of your devices, or one lanowl reaches only by password, is said there, with what to do, and nothing is saved (the backups are written over lanowl's key).
  3. Restart (the banner offers it). Each machine with backup ticked on its page is then backed up monthly, and Back up everything now is on the Security tab.

Edit this page on GitHub.