May 15, 2026
I Built a Personal AI Agent That Runs on My Phone: Here's What I Learned
A Galaxy S21, a local LLM, and cascading inference: building Julian without subscriptions or cloud lock-in.
I wanted an assistant that lived on hardware I controlled — not another chat box renting me back my own habits. So Julian started on a Galaxy S21: local LLM where it could, cascading inference when it couldn’t, and a hard rule against becoming someone else’s subscription.
What I learned wasn’t “phones can run models now.” It was how quickly the product questions show up once the model is no longer the bottleneck: latency you feel in your pocket, privacy you can explain, and the moment cascading to the cloud stops feeling like a cheat and starts feeling like a design choice.
This is the first chapter of that build — why it had to be personal, and what broke first.
That's the setup. Continue on Medium
Follow the work
New essays by email: a list I own, not a Medium follow.