← Field notes

What a Private, On-Device AI Companion Really Means

“Private AI companion” is an easy phrase to put on a landing page. The honest version is specific and checkable, so here it is.

The model runs on your phone

The creature’s words come from a small language model that lives on the device, not an API call to a data center. When it greets you, recalls something, or thinks a half-formed thought out loud, that text is generated locally. Nothing you say is sent off to be processed elsewhere. Small on-device models are more than enough for what this needs: given a few things it remembers about you and what today looks like, write a line or two in character.

The memory stays on your phone

Everything it remembers about you is stored locally. There’s no account holding your history, no sync to a company cloud, no analytics quietly shipping your conversations somewhere. The record of your days that makes the creature yours is held in one place: the phone in your hand.

The mood data never leaves

Once a day the creature asks how you’re feeling, one tap. That’s some of the most personal data an app can hold, so it’s held the only responsible way: on-device, never transmitted, never interpreted or diagnosed. The creature doesn’t analyze you. On a hard day it doesn’t mirror your mood back worse, it just steadies, stays a little warmer, and is present. The architecture guarantees the privacy; the writing makes it kind.

Why “no subscription” belongs in the same sentence

Recurring billing usually exists to pay for servers. Take the servers away, because everything runs locally, and the honest model is a one-time purchase, not a monthly one. Privacy and “no subscription” aren’t two separate selling points here. They’re the same fact seen from two sides: there’s no server, so there’s nothing to bill for and nothing to leak.

That’s the whole idea. Something small and characterful that you raise, that remembers you, and that keeps every private thing about that exactly where it belongs, with you.