lanowl

Running it

Upgrading

Upgrading is pulling the new release: the repository (for the compose file and the examples) and the image. Everything lanowl knows lives in the /state volume and your config folder, so nothing is lost.

cd lanowl
git pull
docker compose -f docker/compose.yaml pull
docker compose -f docker/compose.yaml run --rm lanowl lanowl --check
docker compose -f docker/compose.yaml up -d

With the sandboxed shell, add --profile shell to the pull and up commands, and LANOWL_HOST_IP=<this machine's address> before up unless docker/.env sets it. The releases and what changed are on GitHub; lanowl --version says which one you run.

To run the code you checked out instead of the release (main between releases, or your own change), replace pull with build.

A compose file of your own, copied before the images were published, still names local images (lanowl:latest, lanowl-sandbox:latest): pull then fetches nothing and the old image keeps running. Name the published ones, ghcr.io/alessandromatera/lanowl:latest and ghcr.io/alessandromatera/lanowl-sandbox:latest, or keep using build.

The updater (optional)

Settings → lanowl's version and updates can update lanowl for you. That needs one more container, the updater, which you start yourself:

docker compose -f docker/compose.yaml --profile updater up -d

The risk, in one sentence: the updater holds Docker's socket, which is root on the machine, so it is a container of its own, from the same image, with no network, a read-only disk and no capabilities; lanowl itself never gets the socket and can only tell the updater "update".

What an update does, fixed in the updater's code and not chosen by lanowl: it takes only the containers of its own compose project that are lanowl, sandbox or sandbox-offline and are running; pulls each one's own image (a lanowl built on the machine, with no registry in its name, is refused: update it where it was built); stops there when the images are the same; else replaces the sandboxes first and lanowl last, each new container made from the old one's own configuration. The old ones are kept, renamed -previous, until the new lanowl has run for 3 minutes without restarting and its dashboard answers. Then they are removed. If it does not, the new ones are removed, the old ones come back, and Telegram quotes the new container's last log lines. What happened is in /run/updater/last.json, and on Settings.

Without the updater, update by hand as above (pull, up -d), or pip install -U lanowl and a restart for a pip install.

Before restarting

Run lanowl --check on the new image first. When a setting was renamed or moved, it says so with a ✗ and what to write instead (for example, an old ollama: section is now model:, with name: for the model), and exits with 1, so a script can stop there.

The config folder, read-write

lanowl writes its own files into an empty config folder at a first start, and the dashboard sets secrets too, so the compose file mounts the folder read-write, in one line, instead of read-only with config.yaml and inventory.yaml read-write on their own:

    env_file: [{path: .env, required: false}]    # optional now
    volumes:
      - ../config:/config

If you keep a compose file of your own, make the same change (paths as in yours). Until you do, everything works as before: Settings still writes config.yaml and inventory.yaml, and says secrets.yaml is read-only. required: false needs Docker Compose 2.24 or later (What you need); with an older one, keep env_file: .env and the file.

The model's two shells (its sandboxes) now take LANOWL_HOST_IP from docker/.env or the command line (Settings → The model's shell gives the line); docker/.env keeps working as it was.

There is no dashboard PIN any more: approving on the dashboard is a confirm, behind its login. --check names any actions.pin_sha256, pin_max_failures or pin_lockout_min still in your config.yaml; they do nothing now, so take them out of the file. Devices you watched with the dashboard's Watch button lived in lanowl's state: Settings → Devices offers to move them into inventory.yaml.

What a restart keeps

A restart picks every device up where the last run left it: an outage already under way keeps its start time, an incident you were already told about is not announced again, and the digest does not repeat itself. Messages that could not be sent are still in the outbox and go out when they can. Paused devices stay paused, the model's switch keeps its position, proposals still waiting for you keep their buttons until they expire, browsers stay logged in, and a locked login stays locked.

lanowl is pre-alpha: read the release notes before upgrading, and keep backups.self on, so its own state can be put back.

Edit this page on GitHub.