Alex Finn connects two NVIDIA DGX Sparks and describes their combined 256 GB of memoryAI compute infrastructure is the hardware, facilities, networks, power, cooling, storage, and software used to train and run AI models.. He uses a private Tailscale network and an AI agent to configure the machines and load a local DeepSeek V4.1 Flash modelA local model runs on computing infrastructure controlled directly by the user rather than only through a remote model service..
Alex Finn connects the model to a Hermes agent profile and tests short chat requests, a simple Snake game and a larger three-dimensional game. He reports that the larger game took roughly 20 to 25 minutes and demonstrates the resulting gameplay.
Alex Finn describes local inferenceOn-device inference runs a trained model on the user's device rather than relying entirely on a remote model service. as useful for recurring tasks such as code review and researching business opportunities. He acknowledges that the local model is not a frontier model and says difficult coding work can still benefit from hosted subscriptions.
Watch on YouTube
