Server in Docker¶

Jarvis runs as one container on any Linux machine with Docker: the whole server, the same jarvisd as the other installs, in container mode. On Unraid, use the Unraid template instead; it sets up the same container for you.

Requirements¶

Linux with Docker Engine, and Docker Compose if you use the compose file. The image is for x86-64 (amd64).
Optional: an NVIDIA GPU, with the driver and the NVIDIA Container Toolkit on the host. Without one, everything runs on the CPU.

The image¶

Every release publishes its image as ghcr.io/alexberardi/jarvisd:<tag>, where <tag> is the release's tag (its notes on the releases page name it), and moves ghcr.io/alexberardi/jarvisd:latest to it, pre-releases included. The examples below pin a <tag>, so you choose when to move on; use :latest instead to always pull the newest release. Releases up to rc18 have no image; the first one is rc18.4.

Run it¶

Pick a folder on the host for Jarvis's data, here /srv/jarvis. Then either run it with docker run:

docker run -d --name jarvis --restart=unless-stopped \
  -p 7700:7700 -p 7710:7710 -p 1884:1884 \
  -v /srv/jarvis:/data \
  -e JARVIS_CONTAINER=docker \
  ghcr.io/alexberardi/jarvisd:<tag>

or put this in a compose.yaml and run docker compose up -d next to it:

services:
  jarvis:
    image: ghcr.io/alexberardi/jarvisd:<tag>
    container_name: jarvis
    restart: unless-stopped
    ports:
      - "7700:7700"
      - "7710:7710"
      - "1884:1884"
    volumes:
      - /srv/jarvis:/data
    environment:
      - JARVIS_CONTAINER=docker

JARVIS_CONTAINER=docker is the image's default; it is written out so you can see the container is in container mode, which changes how it is updated.

Moving from the old Docker install and keeping its data? Don't open the setup wizard yet: import your data first.

First run¶

Read the setup token from the container's log:
```
docker logs jarvis
```
It is on the line Setup token: …. It is also in the file /srv/jarvis/setup-token.
Open http://<host>:7710/ and paste the token. That is the setup wizard.
Follow the wizard. It downloads a language model (LLM) and a speech-to-text model; text-to-speech is built in.

Give phones and voice nodes the server's address: http://<host>:7700. They never find a server by themselves, in a container or not.

Ports¶

Port	What uses it
7700	The API: voice nodes, the mobile app and the recipes app
7710	The admin and the setup wizard
1884	MQTT, for voice nodes

Every service lives on 7700 under its own path, so these three ports are all a new install needs; see Ports. Keep the host side of each mapping the same as the container's: clients find the services through /services, which always hands out the container's own ports, so a remap such as 27700:7700 sends phones and nodes to 7700 anyway. If a port is taken, free it on the host.

The data folder and its owner¶

Everything Jarvis keeps (the database, the setup token, models and engines) is in /data, the folder you mounted. Removing the container and running it again keeps your data, and so do updates.

Jarvis runs as user 99, group 100 inside the container. On start, the container makes the data folder belong to that user if it doesn't already. To use another user, set PUID and PGID:

  -e PUID=1000 -e PGID=1000 \

With docker run --user, the container runs Jarvis as that user and changes no owners; the folder must then be writable by it.

Back the folder up with the container stopped (docker stop jarvis). More: Data and backups.

NVIDIA GPU¶

With the NVIDIA Container Toolkit on the host, add two options:

docker run -d --name jarvis --restart=unless-stopped \
  --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=all \
  -p 7700:7700 -p 7710:7710 -p 1884:1884 \
  -v /srv/jarvis:/data \
  -e JARVIS_CONTAINER=docker \
  ghcr.io/alexberardi/jarvisd:<tag>

In compose.yaml, the same is runtime: nvidia and NVIDIA_VISIBLE_DEVICES=all under environment. Set NVIDIA_VISIBLE_DEVICES to one GPU's UUID (nvidia-smi -L lists them) to give Jarvis only that one. The wizard's Hardware step then shows the GPU.

AMD and Intel GPUs are not supported in the container yet.

Updating¶

Jarvis doesn't update itself in a container. Pull the newer image and recreate the container; your data in /data is kept. See Upgrade: In a container.

Coming from the old install¶

To move the old Docker install's data into the container, see Migrating: On Unraid or in Docker. Do it before the setup wizard.