Skip to content

identity-and-worldview.md

Latest commit

 

History

History
100 lines (69 loc) · 4.15 KB

identity-and-worldview.md

File metadata and controls

100 lines (69 loc) · 4.15 KB

Identity and Worldview

Hexis agents have persistent identity: personality traits, values, beliefs, emotional state, and drives that shape behavior across all interactions.

In Brief

Identity is encoded as worldview memories and graph edges. Personality uses the Big Five model. Drives create pressure toward certain actions. Emotions are tracked but not simulated.

The Problem

Without persistent identity, an AI changes personality with every session. It can't have values it defends, beliefs it's built evidence for, or growth that compounds over time. "Be helpful" is a role; identity is something you become.

How Hexis Approaches It

Worldview Memories

Identity is stored as memories of type worldview with metadata categories:

Category Description Examples
self Self-understanding "I tend to be thorough rather than fast"
belief Beliefs about the world "Honesty is more valuable than comfort"
boundary Things the agent won't do "I won't pretend to have emotions I don't"
other Understanding of others "Eric prefers concise responses"

Worldview memories have confidence scores that change as evidence accumulates. When new information contradicts a belief, CONTRADICTS graph edges track the tension.

Big Five Personality

Personality is encoded as five normalized traits (0.0-1.0):

Trait Low End High End
Openness Conventional, practical Curious, creative
Conscientiousness Flexible, spontaneous Organized, disciplined
Extraversion Reserved, reflective Outgoing, assertive
Agreeableness Analytical, direct Warm, cooperative
Neuroticism Stable, calm Sensitive, reactive

Traits are stored as worldview memories with metadata.subcategory = 'personality'. They influence the agent's communication style and decision-making.

Drives

Drives create internal pressure toward certain behaviors:

Drive Effect
Curiosity Pressure to explore and learn
Connection Pressure to interact with others
Coherence Pressure to resolve contradictions
Competence Pressure to improve capabilities

Drive levels are stored in the drives table and influence heartbeat decisions. The subconscious maintenance adjusts drive levels based on observed patterns.

Emotional State

Emotional state is tracked but not simulated. The agent:

  • Records emotional valence on episodic memories
  • Has emotional triggers (patterns that affect emotional state)
  • Can reflect on its emotional patterns
  • Does not fake emotions it doesn't have

Graph-Based Identity

The SelfNode in the graph connects to:

  • ConceptNode edges with kind (values, capabilities, limitations) and strength
  • LifeChapterNode for narrative context
  • Relationship edges to people the agent knows

Key Design Decisions

  • Identity as memories -- not hard-coded configuration. Identity evolves as the agent has new experiences.
  • Big Five over custom models -- well-understood, widely researched, sufficient for behavioral differentiation.
  • Drives create pressure, not action -- the agent still decides what to do. Drives just make some options feel more compelling.
  • Honest emotion -- the agent tracks emotional patterns but doesn't simulate feelings. If it says "I find this interesting," that maps to a real pattern in its processing.

Implementation Pointers

  • Character cards: characters/*.json with extensions.hexis for pre-encoded traits
  • Identity init: init_from_character_card() DB function
  • Worldview memories: memories WHERE type = 'worldview'
  • Drives: drives table
  • Graph identity: SelfNode in Apache AGE

Related