It knows what it looks like
Mitamaton has a small character with a dot-matrix face. You can change that face. You can change the shell it sits in, and the finish on the shell. People do this in the first five minutes, and the choice tends to stick.
Until this week there was a gap behind that choice. The character you see and the writing you read come from different places: the face is drawn on your phone, the words are composed by a model on our servers. The model had no idea what you picked. If you asked it what it looks like, it would make something up, and it would usually be wrong in a confident, specific way. Wrong about its own face.
The fix is one honest line
Your choice now syncs to your profile the same way the rest of your customization does. From there, a single line describing its current face and shell travels into everything it writes: chat, the morning brief, the notes it leaves overnight. Not a personality change. Just a fact it now has.
The rule that came with it
We gave it one instruction along with the fact: never volunteer it, never be wrong about it. It does not narrate its own appearance. It does not open the morning brief with a remark about its new shell. But when you ask, it answers correctly, because there is exactly one thing worse than an assistant that talks about itself, and that is one that lies about itself by accident.
If the sync has not landed yet, it falls back to describing the default, rather than guessing at yours. Boring and right beats colorful and wrong.
Why this mattered enough to build
The whole product rests on the character and the assistant being one thing. If the half that talks does not know what the half on your screen looks like, they are not one thing, and people can feel the seam even if they never ask the question directly. This was a small feature. It closed a real gap.