Skip to main content

The Knowledge Graph

The Knowledge Graph is the "brain" of the Memory Platform. It is a structured representation of everything the platform knows about your subjects, their relationships, and the memories associated with them.

Graph Architecture​

The graph is built using JanusGraph, a distributed, scalable graph database. It consists of:

  • Nodes (Vertices): Representing entities (Users, Orgs, Projects) and MemoryBlocks.
  • Edges (Relationships): Representing the connections between nodes (ABOUT, WORKS_ON, SUPERSEDES).
graph TD
    U[User: Sarah] -- WORKS_ON --> P[Project: Bridge]
    P -- BELONGS_TO --> O[Org: Acme Corp]
    M1[Memory: Prefers PDF] -- ABOUT --> U
    M2[Memory: Status Green] -- ABOUT --> P
    M3[Memory: Headquartered in NY] -- ABOUT --> O
    M2 -- DERIVED_FROM --> D[Document: Report.pdf]

How the Graph Grows​

  1. Explicit Discovery: Relations defined in API calls (e.g., adding a user to a project).
  2. Implicit Extraction: The engine detects mentions of entities in chat transcriptions and creates MENTIONS edges.
  3. Temporal Mapping: As new information arrives, SUPERSEDES edges are created, forming a timeline of knowledge evolution.
  4. Platform Mapping: Your platform's custom entities are injected as nodes, creating a domain-aware network.

The Memory Platform combines the Knowledge Graph with Vector Search (semantic similarity) to provide a complete retrieval system:

FeatureVector Search (Qdrant)Knowledge Graph (JanusGraph)
LogicSemantic similarityExplicit relationships
Search"Find memories about 'budget'""Find memories for Users in Org X"
ContextSingle-node focusMulti-hop traversal
DiscoveryFinds hidden themesFinds structural connections
Best ForNatural language queriesStructural & scoped retrieval

Visualizing the Graph​

The Graph Visualizer provides an interactive UI to explore this data. You can navigate through nodes, expand relationships, and see how knowledge clusters around your primary subjects.

By navigating the graph, you can audit the connections the AI is making and ensure the "context breadcrumbs" it follows are accurate and useful.