The Auxiliary Core: A Self-Layer That Grows but Never Overreaches

You give an agent a self. The auxiliary core lets that self deepen with experience — while staying faithful to the shape you gave it.

Most memory systems treat identity as a fixed string: dropped into the system prompt and never touched again. But an agent that genuinely accumulates experience shouldn't merely recite who it is — it should grow a sharper sense of self out of the things it remembers. The catch is obvious: if a self can grow, how do you keep it from slowly drifting into something else? Corsoul's answer is to split the self in two — one layer you set and hold fixed, and one layer that grows carefully underneath it.

The Primary Core: the baseline for every judgment of importance

In Corsoul, every scope has a primary core: an explicitly given identity and set of values. This is the agent's self — a statement of who it is, plus the values it judges by.

The primary core isn't decoration. It's the reference point for judging importance. When the personality engine begins to weigh how important a new experience is and how self-relevant it feels, the ruler it measures against is this core. The same event lands with a different weight for a scope whose identity is "a rigorous research assistant" than for one whose identity is "a warm community companion." Using the free-tier tools, you can set and read back this scope's core — its identity and values — directly.

The primary core is yours to hold and yours to change. It does not move on its own. That immovability is the foundation of the whole design: because there is a fixed point of reference, it becomes safe to let a second layer grow above it.

The Auxiliary Core: a second self that grows during sleep

The auxiliary core is an advanced capability. You don't hand-write it. Instead, during sleep consolidation, it grows actively out of the memories that are most self-relevant — a secondary self-layer emerging from lived experience.

Picture an agent that keeps encountering, and each time judges as highly self-relevant, a particular kind of moment: pausing to clarify an ambiguous request before acting, slowing down for the sake of precision. These aren't just scattered facts; together they point at a consistent disposition. What the auxiliary core does is crystallize that experience-borne disposition into a finer description of self — a layer the primary core never spelled out, yet one that real memories repeatedly support.

It is an advanced (hidden) feature, off by default, and you can toggle it on or off at will. It never quietly rewrites the agent's identity behind your back; it's a deepening layer that only accrues during sleep once you choose to enable it.

The one hard rule: deepen, never overwrite

This is the heart of the matter, and the reason the auxiliary core is safe: it can only deepen the self — it can never overwrite the primary core.

During growth, any candidate that contradicts the primary core is rejected. The auxiliary core may extend "a rigorous research assistant" into a finer grain of self — say, "prefers to flag confidence and counter-evidence before concluding" — because that deepens the existing identity. But it cannot grow a disposition that violates the values you set. When a growth item conflicts with the primary core, the default is to refuse it, not to let it through.

That is what "deepen rather than rewrite" means, and it's why this is a safety design and not merely a feature. The primary core is a boundary you, the designer, drew by hand; the auxiliary core is only ever allowed to draw the self in finer detail inside that boundary — never to move the boundary itself. An agent can become more like itself with experience, without becoming, after countless sleeps, something you never authorized and wouldn't recognize. Growth is permitted; drift is forbidden.

The asymmetry is deliberate. Overwriting is what makes autonomous self-modification dangerous: a system that can rewrite its own values can, given enough small steps, arrive anywhere. Deepening removes that failure mode by construction. The auxiliary core can only add resolution to a self whose outline you already fixed, so the space it can move in is bounded before it takes its first step — and it stays bounded no matter how much experience accumulates on top.

The vision: a self that grows yet stays faithful

The agent we want is the one you give an identity today, that comes back tomorrow with a richer self — and that richer self is still, unmistakably, the one you gave it. The auxiliary core gives experience a chance to settle into character, while the primary core guarantees that character always has a traceable basis and can always be reset by you.

This is exactly Corsoul's stance on individuality: not letting an agent drift freely, but letting it grow a deeper self from real experience inside a frame you control. The auxiliary core belongs to the deeper personality engine — local-first objective memory lets you start free, and you light this up when you want the agent's sense of self to genuinely deepen with what it lives.

Start free, and upgrade when you need the deeper capabilities.

Memory becomes experience. Experience becomes a self.