npgause/opencpn-kiosk repository overview

⁠OpenCPN Kiosk

A containerized OpenCPN⁠ chart plotter, accessible from any device's browser — no client install needed. Built on Xpra⁠ (remote display, streamed over HTML5) with GPU-accelerated rendering via VirtualGL⁠.

One image works across Intel, AMD, and Nvidia GPUs — or with no GPU at all — the right path is detected automatically at container startup, no build-time choice needed. A separate :pi image covers Raspberry Pi (arm64).

⁠Tags

⁠Quick start

# compose.yaml
services:
  opencpn:
    image: npgause/opencpn-kiosk:x86
    container_name: opencpn
    restart: unless-stopped
    network_mode: host   # needed for NMEA/Signal K mDNS discovery - see "Networking" below
    environment:
      - XPRA_BIND_PORT=14500
      - XPRA_DISPLAY=:100
    devices:
      - /dev/dri:/dev/dri        # Intel/AMD GPU passthrough - omit for CPU-only
    group_add:
      - "993"                    # host's `render` group GID - `getent group render`
    volumes:
      - ./data:/home/ubuntu/.opencpn

docker compose up -d

Then open http://<host>:14500 in a browser.

⁠GPU support

For Nvidia, replace the devices:/group_add: lines above with:

    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

(Requires nvidia-container-toolkit⁠ installed and configured on the host.)

No GPU passed through at all is also fine — both OpenCPN's rendering and Xpra's video encoding fall back to software cleanly on their own.

There are two independent GPU consumers inside the container — OpenCPN's own rendering and Xpra's video encoding — and passing a GPU through doesn't force both to use it:

Set either to false in environment: to force that one consumer to stay software even when a GPU is available.

⁠Configuration

XPRA_BIND_PORT/XPRA_DISPLAY only need changing if something else on the host (another Xpra instance, another X server) is already using the default.

⁠Networking

The example above uses network_mode: host so multicast/mDNS discovery works — NMEA-network devices, Signal K server discovery, etc. Docker's default bridge networking blocks multicast via NAT, which breaks that kind of discovery. If you don't need it and prefer network isolation, use bridge networking instead with a ports: mapping (e.g. "14500:14500") in place of network_mode: host.

⁠Persistent data

OpenCPN's config, routes, and waypoints live in the volume mounted at /home/ubuntu/.opencpn — rebuilding/updating the image never touches it. For chart files, mount your own chart directory in separately, then point OpenCPN's chart directory settings (Options > Charts) at that mounted path — nothing here pre-configures a chart source for you.

⁠USB devices (GPS/AIS)

Pin the device to a stable path with a udev rule on the host (not /dev/ttyUSB0, which can renumber across reboots), then map that stable path through via devices:.