Skip to main content
The Model Orchestrator is responsible for keeping submitted models running and reachable. It is the operational backbone of the Model Nodes.
Diagram showing the Model Orchestrator deploying and monitoring model containers, with a gRPC endpoint for Coordinator access

​
What it does

The Orchestrator:
  • Starts and stops model containers based on on-chain authorizations
  • Ensures one running instance per model — restarts containers if they crash
  • Verifies ownership and permissions by reading the blockchain (in production)
  • Exposes a gRPC endpoint so the Coordinator’s Model Runner Client can discover and connect to models
When a new model is authorized on-chain, the Orchestrator automatically builds a Docker-based runtime, deploys it, and makes the connection details available to the Coordinator.

​
Running locally

The Orchestrator runs as part of your local development stack (via make deploy), letting you simulate a full Model Node on your machine and validate an end-to-end flow before going to production. The Coordinator Platform at http://localhost:3000 provides a UI for submitting and managing local test models — so you can test the complete loop without deploying anything to the blockchain.

Next: Access control

How the Secure Model Protocol authenticates communication between Coordinators and Model Runners.