Learn / Search quality
What is GraphRAG?
Short answer
GraphRAG adds a knowledge graph, a map of the people, organisations, products and other entities in your documents and how they connect, to retrieval. It helps with questions whose answer lies in the relationships between passages rather than inside one passage.
With troveGEN
troveGEN can build a knowledge graph from your documents and use it at search time, so questions that depend on relationships find the passages that connect.
The problem it solves
Chunking breaks documents into separate passages, and similarity search finds passages in isolation. A question like "which suppliers are linked to the delayed project" may need facts from several documents that never appear together. Plain vector search can miss the connections.
How a knowledge graph helps
During ingestion, entities are extracted and linked to the passages that mention them. Entities that appear together are connected. At question time, the system can start from the passages it found, follow the links to related entities, and bring back passages about them. Graphs can also be clustered into communities that summarise topics across a collection.
When to use it
- Multi-hop questions that chain facts across documents.
- Questions about relationships, ownership, dependencies or timelines.
- Broad "what are the main themes" questions over a large collection.
When plain RAG is enough
If the answer sits inside a single passage, vector plus keyword search is faster, simpler and just as accurate. A graph adds ingestion cost and something more to maintain, so add it where relationships genuinely matter.
Key takeaways
- GraphRAG retrieves through relationships between entities, not only through similarity.
- It shines on multi-hop and corpus-wide questions.
- Single-passage questions do not need it.
How troveGEN helps with GraphRAG
troveGEN can build an entity graph from your documents, with co-occurrence links, typed relations and communities, and use it at search time to expand results by shared entity or by following links a few steps. A visual explorer in the console lets you see and check the graph.
What troveGEN provides
- Entity extraction with links between entities that appear together
- Typed relations and communities across a collection
- Search expansion by shared entity or by following links a few steps
- A visual graph explorer in the console
- Permissions re-checked on every expanded passage
Frequently asked questions
Does GraphRAG replace vector search?
No. It complements it. Vector and keyword search find the starting passages; the graph widens the search along relationships.
Is it expensive?
Extracting entities adds processing at ingestion, and the cost depends on whether you use simple extraction or a language model. Use it where the benefit is clear.
Can I see what the graph contains?
You should be able to. Inspectable graphs make answers more explainable, since you can follow the path from a question to its sources.
How does troveGEN help with GraphRAG?
troveGEN can build a knowledge graph from your documents and use it at search time, so questions that depend on relationships find the passages that connect. It provides: Entity extraction with links between entities that appear together; Typed relations and communities across a collection; Search expansion by shared entity or by following links a few steps; A visual graph explorer in the console; Permissions re-checked on every expanded passage.