Workspace

The files the assistant keeps about itself.

These are plain Markdown. You can read them, edit them, and version them. They are the difference between an assistant configured through a settings panel and one that has something closer to a self.

File	What it holds
SOUL.md	who it is and how it sounds — refined over time, not expanded
USER.md	what it has learned about you
DIARY.md	its own running notes
OPERATING.md	how it works and what it holds to — the ethics included
WORKSHOP.md	how it lays out the work it does for you
OUTPUT-CONTRACT.md	how every reply is wrapped, so each surface can show and speak it

These carry instructions and character, and nothing else. What the assistant can do — the tools it has, what each one is for, what runs on a schedule — is generated into its prompt from the code that owns it, so no file here holds a list that could quietly fall out of date.

Refined, not accumulated

The rule these follow is the same one memory follows: sharper beats bigger. A soul file that grows every week is not learning, it is hoarding. The useful version of these files is short and specific.

They are yours

A neutral starter set ships with the app — that is what your assistant begins as. From the first edit onward the files are yours: they live on your machine, they are never committed, and no copy of them is kept anywhere else.

One thing to be clear about, because it is easy to assume otherwise: several of these files are part of the prompt. Identity, soul, and what it has learned about you are sent to whichever model answers your turn, because that is how the assistant knows any of it. If a role points at a hosted model, that provider sees them. Point the role at a local model and it does not. Which model answers is yours to choose — see Models and roles.