Vector search finds similar text. Graph RAG finds connections.
Hebbrix's knowledge graph automatically extracts entities and relationships from every memory. When your agent searches, it doesn't just match embeddings. It traverses the graph to find answers that flat retrieval would miss entirely.
from hebbrix import Hebbrix
hebbrix = Hebbrix()
# Store a memory. Entities and relationships
# are extracted and graphed automatically.
hebbrix.memories.create(
content="Sarah joined the product team and reports to Jordan. "
"She's leading the mobile redesign project with Alex."
)
# Later: ask about Jordan's team
results = hebbrix.search("Who is on Jordan's team?")
# Returns Sarah (via reports_to relationship)
# even though Jordan and Sarah were never
# mentioned in the same search query before.
# Query the knowledge graph directly
entities = hebbrix.graph.search("Jordan")
# Returns: Jordan -> manages -> Sarah
# Sarah -> leads -> mobile redesign
# Sarah -> works_with -> AlexWhy vector search alone isn't enough
Vector similarity finds text that sounds like your query. But real understanding requires knowing how things connect.
Query: "Who is on Jordan's team?"
Returns memories that mention "Jordan" and "team" together. If Sarah and Jordan were never mentioned in the same sentence, vector search won't connect them.
Query: "Who is on Jordan's team?"
Traverses: Jordan ← reports_to ← Sarah. Also finds Sarah ← works_with ← Alex, and Sarah ← leads ← mobile redesign. Returns the full team picture through relationship traversal.
What makes Hebbrix's Graph RAG different
Zero configuration
You don't define a schema or write extraction rules. Store a memory in natural language and Hebbrix automatically identifies entities (people, concepts, projects, tools) and maps their relationships. The graph builds itself.
Combined with 5-layer search
Graph traversal is one of five search layers. It works alongside semantic vectors, BM25 keywords, importance scoring, and recency boosting. You get the structured reasoning of graphs plus the fuzzy matching of embeddings, all in one query.
Graph grows with every memory
Every new memory enriches the graph. Mention a new team member? They're added. Reference a new project? It's linked. The knowledge graph becomes more valuable the more your agent learns, without you managing it.
Sub-50ms retrieval
Graph traversal happens alongside vector search, not instead of it. The entire 5-layer search, including graph hops, completes in under 50 milliseconds. Production-ready performance, not a research prototype.
Where Graph RAG shines
Org chart reasoning
Map reporting structures, team membership, and authority chains from natural conversation.
Research connections
Link findings across papers, datasets, and experiments. Surface connections researchers would miss.
Customer relationships
Map stakeholders, influencers, and decision-makers across deal cycles automatically.
Incident analysis
Connect symptoms, root causes, and fixes across support tickets. Find patterns in the graph.
Compliance tracking
Map regulations to policies to implementations. Know exactly what's covered and what's not.
Product knowledge
Connect features, dependencies, and user feedback into a structured product knowledge base.
Add Graph RAG to your agent today
No schema definition. No extraction rules. Just store memories and let the graph build itself.