Knowledge Graph Explorer
Visualize the entities and relationships automatically extracted from your documents, and explore the knowledge graph with search, group filters, and communities.
A knowledge graph gathers the people, organizations, places, and concepts scattered across your documents — the entities — together with the relationships between them, into a single connected picture. Unlike search that simply finds similar sentences, a graph shows you who is connected to what and how, helping you join scattered facts and see the bigger picture.
Knowledge graphs are built per document group. When you enable the knowledge graph feature on a group and index its documents, the system reads the content and extracts entities and relationships automatically.
Knowledge graph extraction needs an LLM. First register an LLM key in your organization settings and choose a GraphRAG extraction model in RAG settings — only then can you enable the knowledge graph on a group.
Parts of a knowledge graph
A graph is made of four parts. You will meet these same terms throughout the Explorer.
Entities
The key subjects pulled from your documents. The default types are person, organization, place, event, concept, and artifact, and they appear as round nodes in the graph.
Relationships
The links that describe how two entities connect. They are inferred from how often and in what context the entities appear together, and are drawn as lines between nodes.
Communities
Automatic clusters that group closely connected entities. They help you split a large graph by topic and grasp the overall flow through cluster-level summaries.
Organization-wide
Entities with the same name are merged across the whole organization into one unified graph. While exploring, you can narrow the view to just the document groups you want with the group filter.
Building a knowledge graph
A knowledge graph starts by turning on GraphRAG for a document group. When you create or edit a group and enable the knowledge graph option, that group's documents are indexed and entities and relationships are extracted automatically.
Prepare an LLM and extraction model
Register an LLM provider key in organization settings and choose the model used for GraphRAG extraction in RAG settings. This setup must be in place before you can enable the knowledge graph on a group.
Enable the knowledge graph on a group
In document management, create or edit a group and turn on the knowledge graph option. If you like, you can also set which entity types to extract, the output language, and an extra extraction instruction.
Upload and index documents
Adding documents to a graph-enabled group starts indexing. The system reads each document, extracts entities and relationships, and adds them to the graph. You can follow the progress from the documents screen and cancel or retry if needed.
View the results in the Explorer
Once indexing finishes, open the Graph Explorer from the sidebar to browse the extracted entities and relationships visually.
Extraction quality depends on the LLM you use — stronger models tend to find more accurate entities and relationships. If the results feel thin, try a different extraction model in RAG settings.
| Option | Description |
|---|---|
Entity types | The kinds of subjects to extract. The defaults are person, organization, place, event, concept, and artifact, and you can add custom types that fit your domain. |
Output language | The language used for entity names and descriptions. The default, auto, follows the document's language. |
Extra extraction instruction | Free-form notes passed to the LLM about what to emphasize or watch for during extraction. Useful when you want to focus on a specific topic. |
Community summaries | When on, generates a written summary for each community so you can read the overall flow. When off, only the cluster groupings are created automatically. |
Using the Explorer
The Graph Explorer is where you browse the extracted knowledge graph directly. Use the left sidebar for search, filters, stats, and the community list, and click entity nodes on the central canvas to expand their surrounding relationships.
Entity search
Type a name in the sidebar search box to find matching entities. Clicking a result pans the canvas to that node and brings it into focus.
Expand relationships
Clicking a node selects that entity and expands its immediate neighbors. You can hop from one node to the next, following the flow of connections.
Group filter
If the full graph is too broad, pick specific document groups in the sidebar to narrow the scope. Only groups with the knowledge graph enabled appear in the filter list.
Community view
The Community tab lists the automatic clusters; selecting one highlights its member entities. Turn on color mode to shade nodes by community.
When a graph is very large, drawing it all at once can briefly slow your browser. In that case, narrow the scope with the group filter or entity search before exploring.
- Clear the focused entity when the graph canvas has focusEsc
Using it in chat
The knowledge graph is more than an exploration screen. When you connect the graph retrieval tool to an agent, chat can go beyond plain document search and draw on entities, relationships, and community summaries to answer. When an answer is grounded in the graph, you can see the related entities in the answer's sources area and jump straight to the Explorer to dig deeper.
Every graph query is isolated per organization. Search and chat always work only with your current organization's data — no entities or relationships from other organizations are mixed in.
Frequently asked questions
No knowledge graph has been built yet. In document management, enable the knowledge graph feature on a group and index its documents — entities will be extracted and the graph will fill in.
Indexing reads each document to extract entities and relationships, so it takes some time. You can follow the progress on the documents screen, then refresh the Explorer once it finishes to see the results.
No. Every graph query is isolated per organization and shows only your current organization's data. With the group filter you can narrow the view further within that scope.
Adjust the entity types in the group's knowledge graph settings or add an extra extraction instruction to steer the result. Extraction quality also depends on the LLM, so try a better extraction model in RAG settings and re-index.
Communities are automatic clusters of closely connected entities, and a community summary is a written description of the topic each cluster covers. They are generated when you turn on community summaries in the group settings, and they help you quickly grasp the overall flow of a large graph.