VectorMBE Deploy

VectorMBE is a model-based engineering (MBE) runtime built for AI-assisted development teams. It provides a single governed substrate — an OWL-backed graph with vector retrieval, typed constraints, and Model Context Protocol (MCP) integration so engineers, CI pipelines, and LLM tools all reason over the same versioned structure.

This repository is the Docker-only deployment for VectorMBE — no source code required.

---

Requirements

• macOS — Docker Desktop for Mac (Apple Silicon or Intel)
• Windows — Docker Desktop for Windows (WSL2 backend recommended) or Windows Subsystem for Linux (WSL2) with Docker Engine
• Linux — Docker Engine + Docker Compose (plugin or standalone)
• A VectorMBE license key for the official Docker image (request one). Building from source needs no key at all.

Note: radsilent/vectormbe-deploy is public and open source under Apache-2.0, as is VectorMBE itself. Clone it freely. The self-contained quick start immediately below needs nothing but the public Docker image.

---

Quick start without the deploy package

The official Docker image reads the license key at startup; without it the container exits immediately with startup aborted: invalid or missing license. A build from source has no such check — see CONTRIBUTING.

docker run -d \
  --name vectormbe \
  --restart unless-stopped \
  -p 8080:8080 \
  -e VECTORMBE_LICENSE_KEY=your-license-key \
  radsilent/vectormbe:latest

Or, for a persistent setup, write these two files into an empty directory and run docker compose up -d:

# docker-compose.yml
services:
  vectormbe:
    image: radsilent/vectormbe:latest
    container_name: vectormbe
    restart: unless-stopped
    ports:
      - "8080:8080"
    environment:
      - VECTORMBE_LICENSE_KEY=${VECTORMBE_LICENSE_KEY:?Required}

# .env
VECTORMBE_LICENSE_KEY=your-license-key

Open http://localhost:8080, or check it with curl -sf http://localhost:8080/openapi.json > /dev/null && echo up.

Port 8080 is the only port the image exposes — it serves both the UI and the API.

---

macOS

# 1. Download the deploy package
git clone https://github.com/radsilent/vectormbe-deploy.git vectormbe && cd vectormbe
# 2. Configure your license key
cp .env.example .env
# Edit .env and set VECTORMBE_LICENSE_KEY

# 3. Start
docker compose up -d

Open http://localhost:8080 (or the port mapped in your compose file).

---

Windows

Option A — Docker Desktop (recommended)

Open PowerShell and run:

# 1. Download the deploy package
git clone https://github.com/radsilent/vectormbe-deploy.git vectormbe && cd vectormbe
Set-Location vectormbe

# 2. Configure your license key
Copy-Item .env.example .env
# Edit .env and set VECTORMBE_LICENSE_KEY

# 3. Start
docker compose up -d

Option B — WSL2

Inside your WSL2 distro:

# 1. Download the deploy package
git clone https://github.com/radsilent/vectormbe-deploy.git vectormbe && cd vectormbe

# 2. Configure your license key
cp .env.example .env
# Edit .env and set VECTORMBE_LICENSE_KEY

# 3. Start
docker compose up -d

Access the UI at http://localhost:8080 from Windows — Docker Desktop forwards the port automatically.

---

Linux

# 1. Download the deploy package
git clone https://github.com/radsilent/vectormbe-deploy.git vectormbe && cd vectormbe

# 2. Configure your license key
cp .env.example .env
# Edit .env and set VECTORMBE_LICENSE_KEY

# 3. Start
docker-compose up -d

Open http://localhost:8080.

Note: If your system has the newer Docker Compose plugin, use docker compose up -d (space, no hyphen). Both work the same.

---

What's included

• docker-compose.yml — pulls radsilent/vectormbe:latest from Docker Hub
• Caddyfile — reverse proxy for API + static UI
• .env.example — license key, LLM config, and optional feature flags

---

Upgrade

docker-compose pull
docker-compose up -d

Note: Use docker compose (space) if you have the Docker Compose plugin instead of the standalone binary.

---

Logs

docker logs -f vectormbe
docker logs -f vectormbe-caddy

---

Environment variables

Variable	Default	Description
VECTORMBE_LICENSE_KEY	(required)	License activation key, read at startup — the container exits if unset
VECTORMBE_LICENSE_PATH	(unset)	Path to a license file, as an alternative to the key
VECTORMBE_PORT	8080	HTTP port
VECTORMBE_HOST	0.0.0.0	Bind address
VECTORMBE_REQUIRE_TORCH_GPU	false	Set true for GPU-accelerated hosts
VECTORMBE_STARTUP_GRAPH	demo	Preload demo graph on startup (set empty to start blank)
VECTORMBE_ISOLATE_BY_SESSION	false	Multi-tenant isolation — each API key gets its own isolated workspace
VECTORMBE_QDRANT_URL	(unset)	Qdrant endpoint (e.g. http://qdrant:6333) for persistent vector storage
VECTORMBE_LLM_PROVIDER	(unset)	openai, anthropic, or ollama for AI synthesis
VECTORMBE_LLM_API_KEY	(unset)	API key for the LLM provider
OPENAI_API_KEY	(unset)	OpenAI key (shorthand when provider is openai)
ANTHROPIC_API_KEY	(unset)	Anthropic key (shorthand when provider is anthropic)

---

Multi-tenant isolation (shared / demo servers)

When running a shared instance where multiple users connect with different API keys, enable workspace isolation:

VECTORMBE_ISOLATE_BY_SESSION=true
VECTORMBE_ADMIN_KEY=your-admin-key
VECTORMBE_EDITOR_KEY=user-a-key

Each distinct API key gets a fully isolated graph workspace — entities, relations, and vector searches are scoped per key. Users cannot see each other's uploaded models.

---

Optional: Qdrant persistent vector store

For large-scale deployments (50k+ entities), add Qdrant for production-grade ANN with hybrid RRF retrieval:

1. Uncomment the qdrant service block in docker-compose.yml
2. Uncomment qdrant_data in the volumes section
3. Add to your .env:

VECTORMBE_QDRANT_URL=http://qdrant:6333

VectorMBE automatically creates per-kind collections (vectormbe_requirement, etc.) and a cross-kind vectormbe_entities_all collection on first use. Without Qdrant, the in-process HNSW index handles vector search (suitable for most deployments).

---

Troubleshooting

docker: unknown command: docker compose

Your installation uses the standalone docker-compose binary (older style). Replace docker compose with docker-compose in every command:

docker-compose up -d
docker-compose pull
docker-compose logs -f

To install the modern plugin:

sudo apt-get install docker-compose-plugin   # Debian/Ubuntu

---

GPU acceleration (optional)

Requires the NVIDIA Container Toolkit.

Add to docker-compose.yml under the vectormbe service:

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

And set VECTORMBE_REQUIRE_TORCH_GPU=true in your .env.

License

Apache-2.0. See LICENSE.

The VectorMBE source lives at github.com/radsilent/VectorMBE, and contributions are welcome there. Issues with this deployment package — Compose files, Caddy config, the container setup — belong here.