lanowl

Running it

Troubleshooting

Start with lanowl --check, then the log (docker exec lanowl tail -100 /state/lanowl.log). Most problems are named in one or the other, with what to do.

Every device reads DOWN

The host does not allow unprivileged ping (Debian ships it closed):

sudo sysctl -w net.ipv4.ping_group_range="0 2147483647"
echo 'net.ipv4.ping_group_range = 0 2147483647' | sudo tee /etc/sysctl.d/60-ping.conf

If everything went down at once on a lanowl that was working, read the alert: when the router is unreachable too, lanowl says it has probably lost the network itself. Check its own cable, Wi-Fi or VM network first.

The dashboard does not open

  • Is it on? web.enabled: true, and the port: web.port, or LANOWL_WEB_PORT in docker/.env. Without either, the port is 8088.
  • Is lanowl running? docker compose -f docker/compose.yaml ps, then the log.
  • "use the address, not a name" (HTTP 421): you opened it by a host name. Open it by IP address, or list the name in web.allowed_hosts.
  • "Set a password first": the dashboard has no password yet. Send /password to the bot, or set web.password_hash (Logging in).
  • You forgot the password: /password on Telegram sets a new one. If yours is in config.yaml, make a new hash with lanowl --hash-password and restart lanowl.
  • "Login locked": five wrong passwords in a row. Wait 15 minutes, or set a new password with /password, which lifts the lock.

Nothing arrives on Telegram

  • lanowl --check shows whether the token is found and telegram.chat_id is set.
  • Did you send /start to the bot from your account? A bot cannot write to a chat that never started it.
  • "another client is polling this bot" in the log: something else reads this bot's updates (a browser tab on getUpdates, another lanowl). Questions and buttons need lanowl to be the only reader; alerts still go out.
  • No internet: alerts wait in the outbox and arrive, marked delayed, when it is back.

The router refuses the login

The log says "refused the login 'lanowl' (HTTP 401)" once, with what to check: the password in secrets.yaml, and on the router the user's address=, which must include lanowl's host. lanowl does not try again for 15 minutes, or until secrets.yaml changes, because every refused try is a "login failure" line in the router's log.

The owl does not answer

  • The Services list on the Devices tab has a row for the model server: "not answering", or "qwen3:30b not installed" (ollama pull it).
  • Ollama on another machine must listen on the network (OLLAMA_HOST=0.0.0.0), and model.url or LANOWL_MODEL_URL must point at it.
  • Answers that stop half-way, or audits "abandoned after 900s": a slow model for the timeouts (model.request_timeout_s, cadence.llm_max_wall_s). A log line saying the context peak was within 10% of num_ctx means the model was reading a truncated prompt: raise num_ctx.
  • Is it switched off, or is no model set? /model says. With no model: section in config.yaml the owl is off on purpose: Settings → The model sets where Ollama runs.

A device flaps

Weak Wi-Fi, usually. Give its check more patience instead of removing it: {type: icmp, timeout_ms: 2500, count: 3}, or a higher debounce_fails on the device. A device that sleeps by design wants expect_offline. One you switched off wants /pause.

ssh problems

  • "others can read id_ed25519": chmod 600 config/id_ed25519.
  • A device's host key changed (after a reinstall): delete its line from known_hosts_devices in the state volume. One that makes a new key at every boot goes in access.hostkey_any.
  • "needs a login with lanowl's ssh key": the feature needs key: true on that login, and config/id_ed25519.pub in the device's authorized_keys.

The sandboxed shell is off

The owl's shell is offered only while its walls are proven. The log and the Ask tab's "The model's shell" say which test failed: the LAN handshake to shell.canary_lan (pick a service that always answers, such as a server's ssh port), or something the sandbox reached that it must not. Is the sandbox running? Settings → The model's shell gives the command that starts it, with lanowl's address in it.

Times are wrong

Set the time zone: timezone in Settings (the first-run setup takes it from your browser), or TZ in docker/.env, which wins. Sunrise and sunset, the morning jobs, the weekly review and the router's log timestamps all read local time.

Still stuck

Ask the owl: it can read lanowl's own log and records. Or open an issue on GitHub with the --check output and the relevant log lines (they never contain a password or a token).

Edit this page on GitHub.