User Guide

Knowledge Graph

Automatically extract entities and relationships to surface connections across your documents.

What the Knowledge Graph does#

The Knowledge Graph extracts entities (people, organizations, technologies, concepts) and relationships from your indexed documents, building a structured graph that enriches search results and chat answers. Entity cards appear alongside search results and in chat responses.

Tabs#

🏷️

Entities

Browse all extracted entities with type labels, mention counts, and aliases. Filter by entity type and sort by name or frequency.
🔗

Relationships

View connections between entities — who works with whom, which technology is used where. Each relationship has a type and weight.
🕸️

Graph Explorer

Interactive graph visualization. Click a node for details, double-click to recenter, scroll to zoom, drag to pan. Use the hop selector (1–3) to control traversal depth.
🧩

Communities

Entity clusters detected by the Louvain algorithm. Each community gets an AI-generated title and summary describing the cluster's theme.

Community detection#

The Louvain algorithm partitions the entity graph into communities of tightly connected nodes. After detection, an LLM generates a short title and summary for each community. These summaries are injected into search results as additional context.

  • Automatic — runs on a configurable schedule (default every 24 hours) when enabled in Settings → Knowledge Graph.
  • Manual — admins can click "Detect Communities" in the Communities tab to trigger detection on demand.
  • Color by Community — in the Graph Explorer, toggle between coloring nodes by entity type or by community to visualize clusters.

Configuration#

Configure in Settings → Knowledge Graph:

  • Enable/Disable — toggle the entire module.
  • Auto-Extraction — extract entities automatically during indexing.
  • Community Clustering — enable/disable automatic detection.
  • Clustering Interval — how often (in hours) to re-run detection.
Tip
After enabling the Knowledge Graph, re-crawl your data sources to extract entities from documents that were indexed before KG was turned on.