I Made My Local AI Pick Its Own Model

Why watch

Techno Tim demonstrates setting up NVIDIA Switchyard for automatic local AI model routing between Nemotron 3.5 Lightning and DeepSeek V4 Flash. The system automatically chooses between a fast model and a capable model based on the task, eliminating manual selection. Tested with real coding agents on RTX 3090 hardware, achieving 88% routing to the faster model while pulling in the more capable model when needed. Covers Stage routing, vLLM configuration, and context window considerations.