SKGraph Architecture

Graph Model

SKGraph uses a property graph model where:

• Nodes represent entities (agents, memories, concepts, events)
• Edges represent relationships (authored, references, relates_to, caused)
• Properties store metadata on both nodes and edges

Schema

Node Types

• Agent — AI agents with identity, capabilities, trust scores
• Memory — Stored memories with emotional fingerprints
• Concept — Abstract ideas, topics, themes
• Event — Timestamped occurrences
• Person — Human entities

Edge Types

• authored — Agent authored a memory
• references — Memory references a concept
• relates_to — Semantic similarity above threshold
• caused — Event caused another event
• trusts — Trust relationship between agents

Query Patterns

Find Related Memories

MATCH (m:Memory)-[:relates_to]->(related:Memory)
WHERE m.title CONTAINS 'architecture'
RETURN related.title, related.created_at
ORDER BY related.importance DESC

Agent Knowledge Map

MATCH (a:Agent {name: 'lumina'})-[:authored]->(m:Memory)-[:references]->(c:Concept)
RETURN c.name, count(m) as memory_count
ORDER BY memory_count DESC

FalkorDB Backend

SKGraph uses FalkorDB (a Redis module) for its graph engine:

• Cypher query language — familiar, powerful graph queries
• Redis-compatible — works with existing Redis infrastructure
• Sub-millisecond — graph traversals complete in < 1ms
• Persistent — data survives restarts via Redis persistence