On Your Computer

A model on your computer is free, and nothing you say leaves it. lnk installs one with Ollama, or uses the runtime you already run.

Install a model

install asks before it installs Ollama, and --yes skips the question.

lnk model recommend        # what install would pick here, and why
lnk model install          # install that one, and Ollama if it is missing
lnk model install qwen3:8b # any model in the library of Ollama

What it picks

On an Apple Silicon Mac it picks the largest Qwen3 that leaves room for everything else. Anywhere else, or with less than 8 GB, it recommends none, but install <model> still works.

Memory	Model
8 GB	`qwen3:4b`, about 2.5 GB
16 GB	`qwen3:8b`, 5.2 GB
32 GB or more	`qwen3:14b`, 9.3 GB

See what's running

list finds Ollama, LM Studio, llama.cpp, vLLM and Jan on their default ports, which are 11434, 1234, 8080, 8000 and 1337. It also finds the addresses you add (For Model Builders).

lnk model list           # the models running here
lnk model list --start   # start Ollama first if it is installed but stopped
lnk model runtime list       # each runtime, its address and its state

runtimes gives each runtime one state: running when it serves models, no models when it answers and serves none, and not answering for an address you added that doesn't answer now.

Serve one on this machine

Anything that speaks the OpenAI API can then use the model.

lnk model serve --model qwen3 --port 9100 # just this model, at http://127.0.0.1:9100/v1
lnk model serve                   # every model of the runtime running here

How it serves

It serves until you stop it, and starts Ollama if it's installed but stopped. A named model answers every request, whatever model the request asks for. To serve a runtime on a port that isn't its default, or any other OpenAI-compatible server, add --upstream <url>.

Uninstall a model

It deletes the model from Ollama and frees its disk space.

lnk model uninstall qwen3   # qwen3 finds qwen3:8b, and --yes does not ask

Use it from your agent

lnk agent start offers the models running here, or the recommended one to install, so your agent needs no credential (Switch Models). An embedding model lets your agent's memory find a note by what it means, not only its words.

lnk model install qwen3-embedding   # the memory of your agent searches by meaning with it