> ## Documentation Index
> Fetch the complete documentation index at: https://docs.heylinger.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Local and cloud models

> Understand why Linger runs on local models first, when to add a cloud model, and how to switch between them.

Linger runs its agents on a **local model** first. The model runs on your own GPU (graphics processing unit), so your conversations, memory, and files stay on your disk and there is no cloud bill.

A **cloud model** is optional. You add one when your machine cannot run a local model, or when a job needs a larger model than your hardware holds. Linger reaches a cloud model in one of two ways: the subscription of your Runtime, or an upstream model API.

This page compares the two paths and shows how to switch. To see which local model fits your machine, read [Local models](https://docs.heylinger.com/models/local-inference.md).

## How do local and cloud models compare?

Pick the local path when your hardware supports it, and add the cloud path when you need it.

| Attribute | Local model | Cloud model |
| - | - | - |
| Where it runs | On your GPU, through a local gateway. | At the provider of your subscription or upstream API. |
| What you need | Supported hardware. See [Hardware compatibility](https://docs.heylinger.com/hardware/compatibility.md). | A signed-in Runtime subscription, or an API key for an upstream provider. |
| Cost | No per-call cost. | Counts against your subscription or API account. |
| Where prompts go | They stay on your machine. | They go to that provider. |

## How do I use a cloud model?

Use the subscription of your Runtime, or point a Runtime at an upstream model API.

* **Runtime subscription.** Claude Code and Codex use the account you signed in to. Setup can sign you in. See [Choose a runtime](https://docs.heylinger.com/models/choose-a-runtime.md).
* **Upstream API.** Pi calls a model provider with your own API key.

Background work for memory always runs on Codex or Claude Code, with that Runtime's default model. When that Runtime uses a cloud subscription, the conversations it reads go to that provider.

## How do I choose during setup?

The desktop app asks once, during Quick Install. Setup checks the machine first and shows what fits.

**1. Pick the AI runtime**

Choose Auto, Codex, Claude Code, or Pi. Auto takes the choice that setup recommends for this machine.

**2. Turn on the local model**

When the machine can run one, setup names the local model and what it needs: download size, disk, and memory. Select it to run tasks on it. The local model runs behind the Pi Runtime.

**3. Approve the plan**

Nothing downloads until you approve. A model counts as installed only after it answers a real question.

If the machine cannot run a local model, setup says why and offers the Runtimes that use their own sign-in.

## How do I switch between local and cloud?

You switch for each agent, not for the whole install. Every agent has a Computer, a Runtime, and a model. The model list comes from the Runtimes on that Computer, so a local model appears beside the subscription models.

* **For a new agent,** choose the Computer, Runtime, model, and thinking effort when you create it.
* **For an agent you created,** open its direct chat and change the model beside the message box.

Background work does not follow this choice. It always runs on Codex or Claude Code, as described above.

## Keep learning

> [card] [Local models](https://docs.heylinger.com/models/local-inference.md)
>
> See which local model Linger plans for your hardware.

> [card] [Choose a runtime](https://docs.heylinger.com/models/choose-a-runtime.md)
>
> Compare Claude Code, Codex, and Pi, and make one ready.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.