Master AI

Desktop AI agents are invisible. You type into a chat box, the work happens where you can't see it, and you're asked to trust it happened. For people who didn't grow up on computers — and anyone running a business from their phone — that invisibility is the barrier. Busywork doesn't do itself, and an agent you can't watch or stop is an agent you won't use. Tahuti is the visible face of Master AI, a local-first agent framework. He lives on your desktop as a small animated companion and does real work on your own machine — research, summaries, file and app tasks — driven by voice and an Xbox controller through AI Controller. He never acts silently: every consequential action arrives in his one box as a plain-English run card (who, what, where, when, why, how), nothing runs without your approval, and finished work reports back in the same card where you can watch it. In the demo, Tahuti pulls the newest AI headlines and delivers the summary in his card while the desktop keeps working around him. Master AI is model-agnostic — models plug into its routing and can run hosted or fully local, so the user owns the hardware, the accounts, and the data. Built solo, evenings, by an independent developer in Indianapolis. Built with Meta's Muse Spark. Master AI routes each job to the model that fits it; Meta models sit in that routing bench alongside local models that run on the builder's own hardware.