Models across computers

Choose a source model

On the execution computer, open Settings›AI & agents›Orchestration, then Remote agent here. Under Use models from other computers, choose a connected computer, provider and available model. Saved choices from an offline computer remain visible so you can remove them.

The source computer runs inference. The execution computer owns the workspace and tools. Both connections must remain available; choosing a model does not grant access. For the laptop/desktop workflow and data locations, see use desktop models on a laptop.

Share selected models

On the source computer, use Share this computer's models to set its model-access policy. Compatible hosts also offer explicit developer model grants, limits, active-use controls and history. Grant only the models the person needs.

Only models admitted by the source appear as selectable choices. Project context is sent to the source model, and any external provider it uses follows that provider's data policy. Source credentials remain on that computer.

Understand where tools run

The managed remote agent uses the execution computer's permitted project files and supported owned terminals. Its interface describes the actual tool capabilities. External MCP is not supplied by this managed route.

For an owner using a shared local model through the source-computer OpenCode entry, the execution computer's installed OpenCode provides its native tools, modes and MCP configuration. Its normal owner authority applies. Agent tools and instructions explains host tool setup.

When the source is unavailable

Read the source computer card for offline, denied, unavailable or provider errors. Restore the source connection or explicitly select another available model. Revocation or source loss ends affected work; Lyre does not silently substitute a source and replay tools.