Local AI stacks, autonomous robots, on-device computer vision, and one very opinionated NDA analyzer. Everything ships as real code you can fork โ not slideware.
Run Claude Code entirely on a MacBook. 122B-parameter model at 65 tok/s with Google TurboQuant, zero cloud, zero API fees. The same stack underpins AirGap AI โ a commercial service for law firms reviewing NDAs and privileged documents on-device.
Three repos, one MacBook. Brain + ears + hands. Ambient computing that never phones home.
MLX-native Anthropic server + local model lineup (Gemma / Llama / Qwen). The brain of the stack.
Hands-free voice loop โ Apple SFSpeechRecognizer + cloned-voice TTS. Both directions fully on-device.
Local MLX + Chrome DevTools Protocol. Drives a real Brave browser autonomously โ iframes, Shadow DOM, ProseMirror.
macOS screen + facecam recorder with a local HTTP API. Pairs with Claude Code for record-and-send pipelines.
Building toward a modular-robot-parts future โ Lego-like clip-in modules with local AI vision.
1kW autonomous tank platform โ LIDAR, 601-class object detection, iMessage natural-language control, Jetson brain.
Free iPhone app detecting 601 object classes on-device. No ads, no subscriptions, no cloud calls.
Side quests that turned into real results.
Machine-learning tool predicting dopamine-neuron vulnerability. 100% accuracy on a 20-gene signature.
Hybrid chatbot + email support with shared RAG/CAG intelligence, Claude-powered. Runs Divine Tribe customer support.
All of this is free. The code is public, the models are open, and you can run everything on your own Mac without paying me a dime. But if it saved you an afternoon โ or saved your firm from sending a client's NDA to someone else's server โ a few dollars a month keeps me building in the open instead of chasing a day job.
Sponsorships go through GitHub directly to Nice Dreamz LLC. Cancel anytime. No tiers, no paywalled features, no "premium Discord." Just honest support for honest work.
Matt Macosko, working solo out of Arcata, CA. Nice Dreamz LLC. Decades of physical-product iteration before AI accelerated everything into software. Started with a chicken problem. Still figuring it out.
Three ways to engage, depending on who you are: