Choose models on this device or onlineCopy Anchor Link

Lumoswitch separates model connections that run through an online configuration from connections managed by the runtime on this computer. A local routing choice does not mean that inference happens offline: check the connection details before sending private data.

Open Models and add a connectionCopy Anchor Link

1. Open Models in the desktop app.
2. Choose the This device tab for configurations stored on this computer, or Online for account-synced configurations.
3. Under This device, choose Add models to add a provider and select models for a local configuration. To manage provider keys separately, choose Manage sources, then Add provider or edit an existing provider. Enter the API key issued by that provider. For a local model service, use its endpoint and follow the setup prompt. Once a source is available, use New configuration above the local configuration list to create another configuration.
4. Use Online for platform models or existing online configurations. Online BYOK now supports maintaining existing credentials only; adding or replacing a key is no longer supported. Add new keys under This device as described above. To create or change an online routing configuration, use New configuration in the Online configuration panel and select its available models. The server rejects new online keys even if an older client still shows an add-key control.
5. Review and save your changes, then check the selected configuration's status before using it.

The Direct connections section lower in the This device tab manages connections to compatible endpoints separately from the local gateway configurations above it. Check which Agents and workspace features support the selected connection.

Know where a request goesCopy Anchor Link

• A model installed on this computer can process inference locally when the connection points to that local model service. Keep that model service available and use its local endpoint.
• A cloud model configured under This device still sends requests to that provider over the network. Local management or local routing does not make the provider's model run on your computer.
• An Online configuration uses the selected online service and account configuration. Its availability depends on that configuration and the service.

For a local model service such as Ollama or LM Studio, start the service and load the model there first. Then add its reachable endpoint in the local model connection. A local inference endpoint is different from the desktop app's built-in gateway.

Understand billing and the built-in gatewayCopy Anchor Link

When you use your own provider API key (BYOK), the provider may bill that key according to its own terms and usage. A local configuration does not make provider usage free. New keys entered under This device → Manage sources stay on this computer. Previously saved online keys remain managed with your account, with model usage billed by the provider; the online service no longer accepts new or replacement keys. Check the provider's account for charges and limits.

Agent configurations that use the built-in LiteLLM gateway need Lumoswitch to stay running while requests use that gateway. The app may keep running in the background when its window closes. If you fully quit Lumoswitch, the local gateway is unavailable until the app is opened again. A direct native provider configuration has a different lifecycle; check the connection details shown for the Agent.

Confirm that it worksCopy Anchor Link

1. In Models → This device, check the local configuration list and its runtime status. For an online configuration, open Models → Online and check the Online configuration panel.
2. Open Agents and select a compatible connection for an installed Agent, or open Workspace and choose an available model.
3. Send a small request that does not contain secrets or sensitive project data.
4. Confirm that the response arrives and that the expected model connection is selected.

If the connection is unavailableCopy Anchor Link

• For an on-device model, start its model server, confirm the model is loaded, and check that the endpoint is reachable from this computer.
• For a cloud provider using BYOK, recheck the key and provider account limits. Do not paste a real key into chat or documentation.
• If an Agent uses the built-in gateway, open Lumoswitch and check its local runtime status; a fully quit app cannot serve gateway requests.
• If a model is not offered for an Agent or Workspace, choose a compatible connection or review that feature's compatibility information. A saved model connection is not automatically supported by every client.

See Quick start to set up a first connection, Skills for reusable capabilities, or Help for more troubleshooting.