Models Guide
Install or discover a local model, choose a hosted provider, and inspect the model your agent uses. A runtime’s configuration determines where inference runs; provider calls use your credential and the provider’s billing.
-
On your computer, without a Link inference fee.
lnkinstalls a model for you with Ollama, or finds the one you already run. Local inference uses your hardware (On Your Computer). -
A provider's model, with your own credential. Anthropic, OpenAI or OpenRouter. Keys on the wire keep the credential outside compatible harnesses (Model Provider).
-
For model builders. A model you're building gets its own address, for your agent and for a demo (For Model Builders).
lnk uplnk model installlnk model listPick where it runs
lnk model usage, context, and the request actions need the model
namespace of a running agent, as lnk model show does when it names no
model; there, show also says whether the model endpoint is ready, and
which models it has loaded. Outside that context, Link gives a usage message
with the next commands, without offering to report a bug. Use
lnk agent model status for your agent or lnk model list for runtimes
on this computer.
Select your agent's model
Switch your agent between them in one command: Switch Models.
| Costs | Who sees your words | |
|---|---|---|
| On your computer | Your hardware and runtime costs | Determined by the runtime’s configuration |
| A model provider | What the provider bills you | The provider |
| For model builders | Nothing | Your server, and whoever you give the password |
What it needs
You need a Mac or a Linux computer. lnk up adds the module, and
lnk model install picks
a model on an Apple Silicon Mac, and elsewhere you name one. A model on
your own hardware is free.
Where things live
Link keeps two small files, and your models stay in their runtime.
~/Link/Safe/Models/models.tomlholds the addresses you add, and nothing else.~/.lnk/safe/models/facts.tomlholds owner-written model facts. Inspect the merged figures withlnk model show <model>.- Models stay in their runtime, in Ollama's own folder for the ones
installpulls. - Ollama, when
installadds it, is the official one. It's in/Applications/Ollama.app, or~/Applications, on macOS, and a system service on Linux. lnk module remove modelremoves the module, not Ollama or its models.
More about it
Every flag is in lnk model <command> --help. Security
has what's protected and the gaps, and Decisions why it
works this way. When it finds no model: Troubleshooting.
Program API
The model actions API gives agents and sandboxed programs the same request and answer shapes.