Connect a local model

This path is for the person whose machine runs the model. The Noket connector dials out to Noket and carries requests back down that connection, so your machine never accepts inbound traffic and your upstream key never leaves it.

You need a machine that stays on while people use the model, and an LLM server reachable from that machine. Ollama, vLLM, LM Studio, and any OpenAI- or Anthropic-compatible API work.

1. Install and log in

On macOS and Linux:

curl -fsSL https://raw.githubusercontent.com/noket-dev/noket/main/install.sh | bash

On Windows, in PowerShell:

irm https://raw.githubusercontent.com/noket-dev/noket/main/install.ps1 | iex

The installer puts noket and noketd on PATH, installs shell completions, and enables the connector at login. It then starts noket login, which opens the browser for approval. Check the result with:

noket status

The account section reports whether you are logged in and the connector whether it is running. See Install Noket and Log in for the details and manual options.

2. Start your LLM server

Use the server you already run. A few common ones and their default addresses:

Server	Default base URL	Protocol
Ollama	`http://127.0.0.1:11434`	`openai`
vLLM	`http://127.0.0.1:8000`	`openai`
LM Studio	`http://127.0.0.1:1234`	`openai`
Anthropic-compatible	varies	`anthropic`

Load at least one model and note its model id.

3. Add an upstream

noket upstream add

Fill in a name, the protocol, the base URL, and a key if the server needs one. The form probes {base_url}/v1/models and offers the model ids it finds; you can also type an id by hand. The key is read from standard input, never from an argument.

Check what the connector will serve:

noket upstream test your-upstream-name

The probe lists models without running a generation. See Your first upstream for non-interactive examples and the field rules.

4. Start the tunnel

noket up

The connector registers a node — one per machine — and keeps a WebSocket open to Noket. The command prints the process id, the node id and name, and the log path. If the installer enabled the login item, noketd is already running; noket up reports the existing process rather than starting a second one.

5. Create the model

The tunnel carries requests, but people call a model, not an upstream. In the console, open Models, choose Add model, and select Tunnel node. Pick your node, then one of the upstream model ids it reports.

Give the model a slug. Its public id becomes your-username/your-slug. Set a budget now if you plan to share it — you pay for the requests other people make.

See Add a model and Budgets and limits.

To use it yourself, create an API key and call the model id at https://api.noket.dev.
To share it, open the Members dialog and add people or invite them by email. Follow Share your models from step 2.