JARVIS
local-first · apple silicon · MLX · mlx-whisper · ollama · MFCC
A local-first voice assistant running entirely on Apple Silicon — on-device ASR via mlx-whisper, a custom wake-word classifier trained from scratch on raw MFCC features, and a stateful reasoning loop on a local LLM through Ollama. No cloud dependencies: every inference happens on my own machine.
The voice loop already works end-to-end. Next up: the wake-word model, then intent and planning, action execution, and computer vision.
Every layer someone else would import, built and understood by hand. That's the point.
live build log → design notes →