I've been enjoying using the new, native, VS Code integration with local models. I can use LM Studio on my Mac with LM Link to start an OpenAI-compatible server running on my DGX Spark, and plug it in to VS Code for offline/private inference.