Local deployment breakthroughs

Ultra-low-cost hardware (Jetson Orin Nano Super $249, ESP32 $10) enables local inference. Practical guides for CPU-only, Intel iGPU, and constrained RAM setups. Quantization tools (GGUF, SeQTO) and runtimes (Ollama, LM Studio, vLLM) mature. HuggingFace claims fastest inference for DeepSeek V4 Flash and Kimi K3. Security vulnerabilities in llama.cpp highlight risks. New: A critical article argues memory remains the hard unsolved problem for local AI on Macs, challenging MacPaw/Liquid AI partnership hype.