Implemented 5-minute short-term caching for relationship inference:
**store.rs**:
- Added relationship_cache SQLite table
- save_relationship_cache(), get_cached_relationship()
- save_all_relationships_cache(), get_cached_all_relationships()
- clear_relationship_cache() - called on memory create/update/delete
- Cache duration: 5 minutes (configurable constant)
**relationship.rs**:
- Modified infer_all_relationships() to use cache
- Added get_relationship() function with caching support
- Cache hit: return immediately
- Cache miss: compute, save to cache, return
**base.rs**:
- Updated tool_get_relationship() to use cached version
- Reduced load from O(n) scan to O(1) cache lookup
**Benefits**:
- Reduces AI load when frequently querying relationships
- Automatic cache invalidation on data changes
- Scales better with growing memory count
- No user-facing changes
**Documentation**:
- Updated ARCHITECTURE.md with caching strategy details
This addresses scalability concerns for Layer 4 as memory data grows.
Layer 4 provides relationship inference by analyzing memory patterns
and user personality. This is an optional layer used when game or
companion features are active.
Core Philosophy:
- Independent from Layers 1-3.5 (optional feature)
- Inference-based, not stored (computed on-demand)
- Simple calculation using existing data
- Foundation for external applications (games, companions, etc.)
Implementation (src/core/relationship.rs):
- RelationshipInference struct with key metrics:
* interaction_count: number of memories with entity
* avg_priority: average priority of those memories
* days_since_last: recency of interaction
* bond_strength: inferred strength (0.0-1.0)
* relationship_type: close_friend, friend, acquaintance, etc.
* confidence: data quality indicator (0.0-1.0)
Inference Logic:
- Personality-aware: introverts favor count, extroverts favor quality
- Simple rules: bond_strength from interaction patterns
- Automatic type classification based on metrics
- Confidence increases with more data
MCP Tools (src/mcp/base.rs):
- get_relationship(entity_id): Get specific relationship
- list_relationships(limit): List all relationships sorted by strength
Design Decisions:
- No caching: compute on-demand for simplicity
- No persistence: relationships are derived, not stored
- Leverages Layer 1 (related_entities) and Layer 3.5 (profile)
- Can be extended later with caching if needed
Usage Pattern:
- Normal use: Layers 1-3.5 only
- Game/Companion mode: Enable Layer 4 tools
- Frontend calls get_relationship() for character interactions
- aigpt backend provides inference, frontend handles presentation
Layer 3.5 provides a unified, essential summary of the user by integrating
data from Layers 1-3. This addresses the product design gap where individual
layers work correctly but lack a cohesive, simple output.
Design Philosophy:
- "Internal complexity, external simplicity"
- AI references get_profile() as primary tool (efficient)
- Detailed data still accessible when needed (flexible)
- Auto-caching with smart update triggers (performant)
Implementation:
- UserProfile struct: dominant traits, core interests, core values
- Automatic data aggregation from Layer 1-3
- Frequency analysis for topics/values extraction
- SQLite caching (user_profiles table, single row)
- Update triggers: 10+ new memories, new analysis, or 7+ days old
- MCP tool: get_profile (primary), others available for details
Data Structure:
- dominant_traits: Top 3 Big Five traits
- core_interests: Top 5 frequent topics from memories
- core_values: Top 5 values from high-priority memories
- key_memory_ids: Top 10 priority memories as evidence
- data_quality: 0.0-1.0 confidence score
Usage Pattern:
- Normal: AI calls get_profile() only when needed
- Deep dive: AI calls list_memories(), get_memory(id) for details
- Efficient: Profile cached, regenerates only when necessary
Layer 3 evaluates the user based on their Layer 2 memories using the
Big Five personality model (OCEAN), which is the most reliable
psychological model for personality assessment.
Changes:
- Add UserAnalysis struct with Big Five traits (Openness,
Conscientiousness, Extraversion, Agreeableness, Neuroticism)
- Create user_analyses table in SQLite for storing analyses
- Add storage methods: save_analysis, get_latest_analysis, list_analyses
- Add MCP tools: save_user_analysis and get_user_analysis
- Include helper methods: dominant_trait(), is_high()
- All scores are f32 clamped to 0.0-1.0 range
Architecture:
- Layer 3 analyzes patterns from Layer 2 memories (AI interpretations
and priority scores) to build psychological profile
- AI judges personality, tool records the analysis
- Independent from memory storage but references Layer 2 data
