Glossary
RAG vs Graph vs Tabular.
When to use each.
Three retrieval architectures, three different jobs. Vector RAG retrieves passages. GraphRAG traverses relationships. Tabular memory queries structured data. A production AI system uses all three, automatically routed by the question.
"Which RAG architecture should we use?" is the wrong question. Each one is the right answer for a different type of question. The real engineering problem is routing.
What this looks like in practice
Vector RAG: unstructured prose.
Use when the answer lives in documents, transcripts, notes, or free-form text. "What did the procurement playbook say about Net 30 terms?" Vector retrieval finds the passage. Strong on semantic similarity. Weak on exact match, aggregation, or following relationships. Pure vector RAG is the right tool for maybe 60% of the questions your team asks.
GraphRAG: relationships and traversal.
Use when the answer requires connecting entities. "Who owns this account, and what did they decide about pricing last quarter?" The graph traverses the chain across multiple documents. Strong on multi-hop reasoning. Weak on direct passage retrieval and on numeric queries.
Tabular memory: structured numerics.
Use when the answer needs aggregation, filtering, or calculation. "Show me the top 5 customers by revenue this quarter." Structured query runs an actual aggregation against the underlying tables, not a passage search. Strong on numbers and exact match. Useless on prose questions.
Use all three. Route automatically.
We don't pick one. We use all three inside Simply Asking, and the system routes each question to the right architecture based on intent. The user simply asks Lumen. Lumen handles the routing.
Bring your knowledge.Simply ask.
CASA Certified · SOC 2 Infrastructure · GDPR-Aligned · Enterprise Ready