[Feature]: Community Subgraph - Hierarchical Graph Aggregation

[zircote]

Problem Statement

Subcog current knowledge graph is flat - entities and relationships exist at a single level. Zep Graphiti engine implements a three-tier hierarchical structure that enables:

• Community-level insights
• Aggregated relationship patterns
• Scalable graph queries
• Topic clustering

This hierarchy is: Episodes -> Entities -> Communities

Proposed Solution

Implement a community subgraph tier:

Graph Hierarchy:

1. Episode Subgraph: Individual memories/events
2. Entity Subgraph: Extracted entities and relationships (current)
3. Community Subgraph: Aggregated clusters of related entities

Community Features:

• Automatic clustering of related entities
• Community summarization
• Inter-community relationships
• Hierarchical navigation

Benefits:

• "Zoom out" to see big picture
• Faster queries via community-level filtering
• Emergent topic discovery
• Scalable to large graphs

Proposed API:

pub struct Community {
    id: CommunityId,
    name: String,
    summary: String,
    members: Vec<EntityId>,
    parent: Option<CommunityId>,
}

MCP tools:

subcog_communities:
  action: detect  # Run community detection algorithm
  algorithm: louvain
  min_size: 3

subcog_community_get:
  community_id: "comm_123"
  include: ["members", "relationships", "summary"]

subcog_graph:
  operation: visualize
  level: community  # Show community-level view
  depth: 2

Enhanced recall with community context:

subcog_recall:
  query: "database decisions"
  community_boost: true  # Boost results from same community

Alternatives Considered

• Flat graph only (current) - does not scale
• Manual tagging for grouping - tedious
• External graph analytics - lose integration

Additional Context

• Zep hierarchical graph: https://arxiv.org/abs/2501.13956
• Community detection algorithms: Louvain, Leiden
• Enhances existing graph capabilities

Breaking Change: No
Priority: Nice to have
Contribution: Yes, with guidance