What Gets Extracted
When a file is processed, an AI model reads the content and identifies:- Entities — Named things like people, companies, products, locations, dates, financial figures, and domain-specific concepts
- Relationships — How entities relate to each other, such as “works at”, “acquired by”, “depends on”, or “located in”
Entity Types
The knowledge graph recognizes several categories of entities:| Type | Examples |
|---|---|
| Person | Employee names, executives, authors, contacts |
| Organization | Companies, departments, agencies, vendors |
| Product | Software, services, hardware, offerings |
| Location | Cities, countries, offices, regions |
| Concept | Technical terms, methodologies, standards |
| Date/Event | Specific dates, milestones, deadlines |
| Financial | Revenue figures, budgets, valuations |
Exploring the Graph
Open the knowledge graph from the Knowledge Hub in the sidebar. The default view shows a network visualization with nodes representing entities and lines representing relationships.Use the search bar at the top of the graph view to find a specific entity by name. The graph will center on that node and highlight its immediate connections.
Clicking any node opens a detail panel showing the entity name, type, the source documents it was extracted from, and all of its relationships. You can click through to the original file to see the entity in context.
The neighborhood view expands outward from a selected node to show all entities within one or two relationship hops. This is useful for understanding the full context around a person, company, or concept — who they work with, what projects they are associated with, and which documents mention them.
Select two entities and use the shortest path feature to discover how they are connected. RelayHub calculates the shortest chain of relationships linking the two nodes, even if the connection spans multiple documents. This can reveal non-obvious relationships — for example, two companies connected through a shared vendor mentioned in separate contracts.
Relationship Visualization
Edges in the graph are labeled with the relationship type. The visualization uses color and thickness to convey information:- Edge labels describe the relationship (“reports to”, “supplies”, “references”)
- Node size reflects how many connections an entity has — heavily referenced entities appear larger
- Clustering groups related entities together spatially, making it easy to spot organizational structures or topic clusters
Filtering the Graph
For large document libraries, the full graph can be dense. Use the filter controls to narrow what is displayed:- By entity type — Show only people, or only organizations, to simplify the view
- By source file — Limit the graph to entities from a specific document or set of documents
- By scope — Filter to workspace-level or company-level knowledge
How the AI Uses the Knowledge Graph
The knowledge graph is not just a visualization tool. When you ask questions in chat, the AI can query the graph to find relationships and context that go beyond simple text search. For example:- “Who are the key contacts at Acme Corp?” — The AI finds all Person nodes connected to the Acme Corp organization node
- “What products are mentioned alongside our Q3 revenue report?” — The AI traverses relationships from the report to connected product entities
Knowledge graph data is rebuilt when you reprocess a file. If you delete a file, its entities and relationships are also removed from the graph.