Complete your first desktop task
Start with one small task, then add Skills, MCP or QQ playbooks. You do not need to learn every API or routing term first.
1. Install and open
Get the right package for your operating system and chip from the official download page. If detection is uncertain, use its other-version selector. Install and open Lumoswitch, then follow the app's sign-in prompts. If account protection asks you to verify a phone number, complete that step to continue; see Help if the check fails.
When upgrading, retain the existing application data and use the same account. Do not first uninstall and clear its data; see help and updates.
2. Choose a model connection
Open Models in the sidebar. The top tabs are This device and Online:
| What you have | What to choose |
|---|---|
| You want a platform-provided model | Open Online and choose or create an available configuration |
| Your own provider API key | Open This device and choose Add models; use Manage sources to add or edit a provider API key, then select models |
| A model service such as Ollama already on this computer | Start that service, then add its address and model under This device |
Save and select the configuration. Merely browsing local or online tabs does not select the workspace's active connection; check the current selection.
Cloud models configured locally still incur provider charges. Platform models follow platform billing. Connection checks and the verification request below can consume usage. See model connections for details.
3. Open a workspace and send a small task
Open Workspace and select an empty practice folder. At the bottom of the composer, choose a model configuration first, then a model within it. Start outside projects containing important files.
For this first check, select Default permissions and leave Auto approve and computer use off. Asking the model not to change files is a task instruction, not a permission control. You can reject file or command approval requests during this simple question-and-answer check.
For example:
Suggest three easy home-cooked dishes and list their main ingredients. Reply only in chat; do not modify files or run commands.
Check the model selection and run status. If an approval request appears, review what it proposes before deciding. This first check does not require installing external tools.
4. Verify success
A complete reply with the run no longer active or awaiting confirmation completes this check. A “configuration saved” message alone does not verify the entire connection.
If the app asks you to reconnect or confirm the current run's state, follow the workspace guide. Do not repeatedly resend an uncertain task or delete its working folder to troubleshoot it.
Already use an Agent?
In Agent, select an installed tool, preview the proposed model configuration, confirm it and reopen the Agent as instructed. Send a small request there to verify it. Installation and configuration support varies by tool; see Agent management.
Next steps
- Add reusable capabilities: Skills.
- Configure external tool servers: MCP.
- Chat in QQ: Big Fish.
- Customize your companion: Desktop pets.
- Resolve a problem: Help and updates.
Desktop user guide
Use the Lumoswitch desktop Agent console and AI toolbox: connect models, configure Agents, add Skills and MCP, start a workspace task, and find help.
Choose models on this device or online
Add an online or on-device model connection, understand where requests run and who bills them, then confirm the runtime is ready before connecting an Agent or Workspace.