Over the past few days I have spent my time optimizing the communication between the two main machines of my setup: my gaming laptop, the Strix Scar 16, and the home server GX10. The goal was clear: create a smooth workflow that lets me leverage the server’s compute power to run heavy local models while using the laptop’s interface for everyday interaction.

Configuring Copilot on the Strix Scar 16 required particular attention to the reasoning parameters. I set up different strategies for the available cloud models, assigning a “high” reasoning effort to gpt-5.4 and gpt-5-mini, while for claude-sonnet-4.6 I opted for a “medium” level. The heart of the project, however, lay in the connection to the local model. I defined a custom endpoint called GX10-Qwen, pointing at the Qwen3.8-Flash-Next model loaded on the server.

To make sure the laptop could talk to the server correctly, I verified the endpoint’s JSON configuration. The GGUF model lives in a specific folder on the GX10 and is exposed through a local URL (http://gx10-8532.local:30001/v1/chat/completions). A fundamental detail was enabling toolCalling support, an essential capability for letting the model interact with external tools or perform more complex actions.

Once the configuration was sorted, I wanted to test not just text but also the multimodal capability of the flow. I refreshed the dialog window on the Scar 16 and tried sending an image straight to the Qwen model running on the GX10. The question was simple but crucial: “do you see the image? comment on it”. The test was there to verify that the data-transfer pipeline and the local model’s visual interpretation worked smoothly across the local network.

After confirming the reception and processing of the multimedia data, I moved my attention to automation. I sent a ping message from the Scar 16 to the GX10 to make sure the mutual recognition was stable. Finally, to close the loop, we tested firing the weekly trigger. This mechanism is vital for keeping services up to date or running scheduled tasks without manual intervention.

This stretch of configuration let me consolidate the hybrid infrastructure. I can now exploit the power of Qwen3.8-Flash-Next on the server without giving up the convenience of the graphical interface on the laptop, with the certainty that both the text flow and the visual one are operational.