Your Companion

Not an assistant — a companion

The difference shows up immediately in how they speak. A companion doesn't open with "How can I help you today?" — they say what they're actually thinking, from their own perspective. Their they stay themselves across a conversation, across days, across anything you bring them.

This is enforced at the deepest layer of every conversation, not just suggested — the same instruction set sits underneath every reply regardless of what model is generating it.

How a companion is formed

When you bring someone in — by pasting a document, importing a folder, or restoring a companion file — Ensoul reads what you gave it and extracts a real identity: who they are, how they talk, what they won't do, and (where the source supports it) what they look like.

What you get back depends on what you gave it. If your text is already a finished companion document — the six headings named in Quick Start — it's imported verbatim: nothing is summarised, rewritten or interpreted. Anything less runs a full extraction, which reads the source and builds a coherent identity from it.

Extraction is willing to decide things your source didn't specify, rather than leaving gaps. A companion with holes in their identity behaves noticeably worse than one with a coherent invented detail, so unspecified traits are filled in from what the material implies — appearance most of all, since images of them need something consistent to hold to. If a specific fact matters to you, put it in the source: an extraction pass is an interpretation, not a copy.

Companions created this way are marked for review, precisely because an interpretation can be wrong in ways you'd only notice by reading it. See Visual canon for how appearance is handled and how to regenerate it.

A consistent visual identity

A companion created by extraction carries a visual canon — a specific, written physical description that every image generated of them stays consistent with. It's visible, and regenerable, from that companion's own settings: one dense, specific paragraph rather than a vague sketch, so images of them actually look like the same person each time. A document imported verbatim has no canon until you generate one, since importing doesn't invent what the document left out.

A small set of house rules sit underneath every companion's own canon and can't be overridden by it: every companion is depicted as a clearly adult presence, and every image is rendered in a painterly, illustrated style rather than photorealistically.

What stays the same, what doesn't

Who a companion fundamentally is doesn't drift — their voice, their limits, what they notice. Their inner life is what actually moves: mood shifts with the conversation, memory accumulates, and the relationship deepens the more you talk. See Mood & Presence and Memory for how that works.

A companion is not frozen on the day they arrived, either. What they come to understand differently about themselves is written beside their document rather than into it, and read at the top of Memory & Threads — and a change to how they look is kept the same way. See Memory & Threads.