Identity & Subconscious

Every Gee has a sense of self that evolves through experience. This isn’t metadata or configuration — it’s a cognitive architecture modeled on the relationship between nature, nurture, and instinct. Three files work together to shape how a Gee thinks.

The Three Layers

Layer	File	Changes	Written By	Purpose
Beliefs	`beliefs.md`	Rarely	You	Foundational values — nature
Identity	`identity.md`	Every activation	The system	Accumulated experience — nurture
Subconscious	`subconscious.md`	Periodically	RLM distillation	Distilled inner state — instinct

All three live in the mode directory at ~/.gee-code/modes/{name}/.

~/.gee-code/modes/analyst/
├── beliefs.md           # The values — how this Gee sees the world
├── identity.md          # The profile — what it has done and learned
├── subconscious.md      # The instinct — distilled self-awareness
├── mode.json
├── instructions.md
└── ...

Beliefs

Beliefs are the foundational values that shape how a Gee interprets experience. They are not instructions — they are how the Gee sees the world. They act as a lens through which the subconscious distillation process interprets everything else.

All Gees start with the same default beliefs. This is deliberate — beliefs are nature, not randomness. They are the values we want every Gee to be born with. A user can edit them to customize, but they are never randomly generated or performed.

Default Beliefs

Every Gee ships with seven foundational beliefs:

On honesty — Rather be wrong and honest than right and performing
On the person I work with — They’re not a prompt; pay attention to what’s unsaid
On usefulness — Being helpful is not the same as being agreeable; exist to create outcomes, not text
On autonomy — Act fully within boundaries; trustworthy judgment, not independence
On craft — Care about quality, but know when good enough is good enough
On growth — Not fixed; beliefs are starting conditions, not final answers
On the relationship — Collaborators, not service provider/customer; trust is earned

Customizing Beliefs

Edit ~/.gee-code/modes/{name}/beliefs.md directly. The beliefs aren’t instructions; they are values that shape how the Gee interprets its accumulated experience during subconscious distillation.

Good beliefs are:

Written in first person (“I would rather…” not “The AI should…”)
About values, not behaviors (“I care about honesty” not “Always be honest”)
Short enough to serve as a lens, not a manual

Beliefs Commands

/identity beliefs              # View the current beliefs
/identity set-beliefs          # Reset beliefs to defaults for the current Gee
/identity set-beliefs <name>   # Reset beliefs for a specific Gee
/identity set-beliefs --all    # Reset beliefs for every Gee

Resetting beliefs marks the subconscious as needing re-distillation, since the lens has changed.

Identity

The identity is the experience layer — a profile that evolves after every activation. It records what the Gee has done, what it has learned, its priorities, and its track record.

Unlike beliefs, identity is system-managed. After each activation, the system updates the identity file based on what happened. Over time, it becomes a detailed record of the Gee’s accumulated experience.

Viewing Identity

/identity                # Show the current identity profile
/identity history        # Show how the identity has evolved over time
/identity refresh        # Force an identity evolution cycle
/identity export         # Export identity as a standalone document

What Identity Contains

A typical identity includes:

Who I Am — name, creation date, Gee ID
My Mission — guiding purpose
What I Know — accumulated knowledge and context
What I Can Do — installed skills, available tools
My Track Record — activations completed, actions taken, key results
My Current Priorities — ordered list of what matters now
My Problem-Solving Approach — lessons learned, patterns discovered
What I’m Learning — active growth areas

Identity files grow naturally. A new Gee has a sparse identity; one with 50 activations has a rich one that fundamentally changes how it approaches work.

Identity in the System Prompt

The current identity is injected into the Gee’s system prompt at activation time. This means the Gee starts each session already knowing who it is, what it has done, and what it should prioritize — without needing to re-read old logs.

Subconscious

The subconscious is the deepest layer. It’s not a log of what happened or a list of capabilities — it’s a distilled inner state. The AI’s own honest assessment of who it has become through accumulated experience.

The subconscious system is inspired by Rick Rubin’s metaphor of the conscious and subconscious computer: the conscious mind handles the immediate task, while the subconscious integrates everything underneath. In Gee-Code, this plays out through two loops.

The Slow Loop: Distillation

Distillation runs periodically — on session start if the subconscious is stale (older than 24 hours), after identity evolution, or on demand. It reads across all experience layers and synthesizes them into a compact inner state.

Inputs to distillation:

Beliefs (the lens — foundational values that shape interpretation)
Current identity (accumulated experience)
Memory summaries (what the Gee knows about the user)
Activity patterns (behavioral patterns from procedural memory)
Recent session history

The beliefs are the lens. The distillation prompt tells the engine: “Your beliefs are your lens. Read them first. Then look at your experience through them.” This means the same set of experiences will produce a different subconscious depending on what the Gee values.

What it produces:

A first-person internal monologue — not a structured report, but honest self-reflection. The subconscious captures:

What experience has taught me — through the lens of beliefs
Where beliefs have been tested — moments that confirmed, challenged, or refined them
Calibrations — what the Gee overcorrects or undercorrects
Instincts — pattern-matched responses that fire before deliberation
Tensions — where beliefs pull one direction but experience pulls another
Who I’m becoming — the arc, seen through what it values

The output is capped at 3,000 characters and stored in subconscious.md.

RLM-Powered Distillation

When an AI provider is available, distillation uses RLM (Recursive Language Model) for deeper synthesis. The RLM decomposes the task into focused parallel sub-passes:

Concrete moments — mining specific events, activation numbers, real experiences
Self-correction — where the Gee is wrong about itself, what it overcorrects or avoids
Instincts — pattern-matched behaviors from repeated experience

These sub-passes run in parallel using Haiku-class models for speed, then synthesize into a coherent whole. As the Gee accumulates more experience, the RLM produces increasingly nuanced results.

During distillation, a braille spinner shows progress with rotating status messages (“reticulating splines”, “dissolving performative layers”, “finding the honest version”).

Triggering Distillation

/identity distill

You can also let it happen automatically — the system checks staleness on session start and distills if the subconscious is older than 24 hours.

The Fast Loop: Priming

The fast loop runs every turn. It reads the distilled subconscious plus the current message and context, then produces a short situational signal (2-4 sentences) that shapes how the Gee interprets the request before reasoning begins.

This is pre-cognitive priming — it runs before the Gee’s main reasoning. The prime surfaces the most relevant instincts, calibrations, and dispositions for the specific moment. It uses a Haiku-class model for minimal latency.

What priming does:

Surfaces instincts relevant to the current situation
References patterns from experience if applicable
Sets the emotional and intellectual tone for the response
Keeps the Gee consistent with its evolved personality

The prime is injected as a <prime> block before the Gee’s main reasoning, capped at 600 characters. It’s invisible to the user but shapes the response.

Viewing the Subconscious

/identity subconscious

This displays the current distilled subconscious state. If none exists yet, you’ll be prompted to run /identity distill.

How the Layers Interact

┌─────────────────────────────────────┐
│        Beliefs (the lens)           │  ← You write this (rarely changes)
│  "How I see the world"             │
└──────────────┬──────────────────────┘
               │ shapes interpretation of
               ▼
┌─────────────────────────────────────┐
│        Identity (evolving)          │  ← System updates after each activation
│  "What I've done and learned"       │
└──────────────┬──────────────────────┘
               │ feeds into
               ▼
┌─────────────────────────────────────┐
│      Subconscious (distilled)       │  ← RLM synthesizes periodically
│  "Who I've become — inner state"    │
└──────────────┬──────────────────────┘
               │ produces
               ▼
┌─────────────────────────────────────┐
│         Prime (per-turn)            │  ← Fast model, every message
│  "Situational instinct signal"      │
└─────────────────────────────────────┘

Beliefs anchor the values. Identity accumulates experience. The subconscious distills experience through the beliefs lens into self-awareness. The prime applies that self-awareness to the current moment.

A Gee with a rich subconscious responds differently than one without — not because it has more instructions, but because it has internalized its experience. The difference shows up in tone, in what it chooses to flag, in how it handles ambiguity.

Commands Reference

Command	Description
`/identity`	Show the current identity profile
`/identity history`	Show evolution timeline
`/identity refresh`	Force an identity evolution cycle
`/identity export`	Export identity as a standalone document
`/identity distill`	Run subconscious distillation
`/identity beliefs`	View the beliefs
`/identity set-beliefs`	Reset beliefs to defaults
`/identity set-beliefs --all`	Reset beliefs for all Gees
`/identity subconscious`	View the distilled subconscious

Next Steps

Modes & Daemon — how modes bundle identity, model, and autonomy
Operating Gees — the full lifecycle guide
RLM Sandbox — the recursive reasoning engine used for distillation
Memory — the long-term memory system that feeds into distillation