Offline first.
Everything runs on local models. No API bill, no data leaving the machine.
Developer
Local-first AI.
On your own Mac.
I build tools that keep working offline: voice assistants, multi-model councils, and the plumbing that keeps a hundred background agents healthy.
Open source
Built here.
Python · macOS
Jarvis
Voice-first Mac assistant that keeps working offline. Ollama brain, local Whisper wake word, Piper voice, Mac control.
View on GitHub › Python · Ollama ollama-councilTwo local models cross-check, a third breaks ties. Meaning-based agreement, FIFO queue, injection screen, secret redaction. Zero dependencies.
View on GitHub › Python · LLM verification triple-checkStop local LLMs making up numbers. Python does the math, two experts explain it, a different model family verifies every figure.
View on GitHub › Python · macOS launchd-doctorYour LaunchAgent exits 78 with no output, or runs 84 times a day. Find and fix the launchd bugs that fail silently.
View on GitHub ›Principles
Private by default.
Verify, don't trust.
Numbers are computed, not generated, and checked by a second model family.
Fail loudly.
Silent failures get detected, healed and logged, with circuit breakers for anything that keeps breaking.