Run the connector

The Noket connector is the part that runs on the machine with your models. It ships as two binaries:

• noket — the command-line client: log in, manage upstreams, start and stop the tunnel, and inspect what is happening.
• noketd — the background process that keeps the tunnel open, probes upstreams, and forwards inference requests to them.

You need the connector only if Noket should reach a model on your hardware or network. Sharing a public API you already have credentials for does not need it.

The path

1. Install Noket — binaries, PATH, completions, and the login item.
2. Log in — approve the CLI in the browser and store the token on this machine.
3. Add and manage upstreams — describe each LLM API the connector can reach.
4. Run noket up and create a model in the console bound to this node.

If you already have a server running, the local model guides cover Ollama, vLLM, LM Studio, and Anthropic-compatible servers.

How the pieces relate

A node is one machine running the connector. Its node id is created on the first start and stays with the machine. An upstream is an LLM API reachable from that machine. A model in the console can be bound to one upstream model id on one node; the tunnel connects the two.

The connector holds the upstream keys in config.toml on that machine. Noket stores them encrypted for external models and never sends a tunnel model’s upstream credentials to the browser.

Everyday commands

The CLI reference documents every command and flag.

Keeping it running

The installers enable the connector at login, so it starts when you sign in and restarts if it fails. If you install with --no-autostart, you can still run noket up yourself or set up supervision later; see Run as a service.

Next

• Use the terminal UI
• How the tunnel works
• Security and privacy