wavyrai/macspark

MacSpark is a written guide, not a piece of software you install and run. It is aimed at people who own an NVIDIA DGX Spark, a specialized computer built for running AI models, and who want to control and use it from a Mac. The repository explains, in careful steps, how to connect the two machines safely and what to set up once they are connected. The guide is explicit that it is documentation and reviewed examples rather than an automatic installer. Following it involves running commands yourself that can change networking, authentication, installed packages, containers, and remote access settings, so the author repeatedly stresses reading each part first and keeping a way to recover the machine through its console if something goes wrong. Topics covered include setting up a private, secure network connection between the Mac and the DGX Spark using Tailscale or verified SSH, checking the machine's setup before making changes, running an AI model locally on the DGX Spark through a tool called vLLM in a way that other programs can talk to like a standard AI service, and optionally routing between local and other AI models through a gateway. It also covers convenience features such as sharing files, using a remote code editor, and remote desktop access, along with how to keep these services running over time and how to safely update or roll them back. The recommended path is to go one step at a time: confirm the basic hardware setup works, connect the network, bring up one AI model service, test it, and only then add extra convenience features one by one. The project explicitly does not cover exposing the DGX Spark to the public internet or automatically installing operating system updates. The repository also includes small example configuration files for the tools it discusses, plus two optional automated helpers meant to assist with planning a setup or diagnosing connection problems. It is released under the MIT license.