Multi-Machine

For larger deployments, run the coordinator on one machine and workers on as many machines as you want.

Architecture

┌─────────────────┐
│   Coordinator    │ ← Dashboard, job queue, worker registry
│  (server-1:7420) │
└────────┬────────┘
         │
    ┌────┼────────────┐
    │    │             │
┌───┴──┐ ┌───┴──┐ ┌───┴──┐
│Worker│ │Worker│ │Worker│
│Ollama│ │Claude│ │Laptop│
└──────┘ └──────┘ └──────┘

Coordinator setup

On the coordinator machine:

modelreins coordinator start --host 0.0.0.0 --port 7420

Worker setup

On each worker machine, point at the coordinator:

export MODELREINS_API_URL=http://coordinator-host:7420
export MODELREINS_API_KEY=your-shared-key

modelreins worker start --name $(hostname) --provider ollama

Security

• Always use MODELREINS_API_KEY to authenticate workers.
• Use HTTPS (reverse proxy) for coordinator access over the internet.
• Workers only need outbound access to the coordinator and their provider APIs.

Scaling

Add workers at any time — they register automatically. Remove workers by stopping the process. The coordinator requeues any in-progress jobs from disconnected workers.

Companion App on spare machines

The easiest way to add machines to the mesh:

npx @mediagato/modelreins-companion install

See Companion App for details.