Self-hosted

Deploy VectorMBE

One Docker image, on your own infrastructure. Nothing leaves your network.

bash

docker run -d --name vectormbe \
  -p 8080:8080 -p 8081:8081 \
  radsilent/vectormbe:latest

Then open http://localhost:8080 and enter your license key.

Requirements

Docker, on any platform that runs it
Ports 8080 (UI and API) and 8081 free
A VectorMBE license key — each key activates one production instance
CPU inference works out of the box; a GPU is optional

Quick start

1 Pull the image

bash

docker pull radsilent/vectormbe:latest

2 Run it

bash

docker run -d \
  --name vectormbe \
  -p 8080:8080 \
  -p 8081:8081 \
  radsilent/vectormbe:latest

Port 8080 serves the UI and API. Port 8081 is the MCP endpoint.

3 Enter your license

Open http://localhost:8080 and enter your license key on login to launch your instance. Building from source or running the desktop app instead? See the MBSE install guide.

Platform notes

The commands are the same everywhere Docker runs. Only the prerequisite differs:

macOS — Docker Desktop, Apple Silicon or Intel
Windows — Docker Desktop with the WSL2 backend, or WSL2 with Docker Engine
Linux — Docker Engine and Docker Compose

To clone the deploy package and bring it up with Compose:

bash

git clone https://github.com/radsilent/vectormbe-deploy.git vectormbe
cd vectormbe
docker compose up -d

Docker Compose

For a production server that should come back after a reboot, save this as docker-compose.yml:

yaml

services:
  vectormbe:
    image: radsilent/vectormbe:latest
    restart: unless-stopped
    ports:
      - "8080:8080"
      - "8081:8081"

Then docker compose up -d.

Reverse proxy

The included Caddy config serves the UI and proxies API calls from one domain, with automatic HTTPS and no JavaScript changes:

Caddyfile

yourdomain.com {
  handle_path /api/* {
    reverse_proxy localhost:8080
  }
  root * /var/www/vectormbe
  file_server
}

It ships in docker-compose.prod.yml.

GPU acceleration

Optional, and requires the NVIDIA Container Toolkit.

bash

docker run -d \
  --name vectormbe \
  --gpus all \
  -e VECTORMBE_REQUIRE_TORCH_GPU=true \
  -p 8080:8080 \
  -p 8081:8081 \
  radsilent/vectormbe:latest

The Compose equivalent:

yaml

services:
  vectormbe:
    image: radsilent/vectormbe:latest
    restart: unless-stopped
    ports:
      - "8080:8080"
      - "8081:8081"
    environment:
      VECTORMBE_REQUIRE_TORCH_GPU: "true"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

Verify it is running

bash

docker ps --filter name=vectormbe
docker logs --tail 20 vectormbe
open http://localhost:8080  # macOS

On Linux use xdg-open http://localhost:8080; on Windows, start http://localhost:8080.