Examples

Queries

Retrieving information using Standard, Unified, or Vector Bypass modes.

I/O Module

Query Modes

Not all questions are equal. Sometimes you need deep reasoning (Graph), sometimes you just need speed (Vector).

Method	Best For	Mechanism
execute()	Complex QA	Graph Traversal + Vector + LLM
execute_unified()	Multi-KG Search	Searches across multiple graphs & memories
vector_bypass()	Speed / Simple Similarity	Vector Search Only (No Graph)

Code Example

question = "How to prepare for a 5K run?"

# 1. Standard Graph RAG Query (Detailed)
# Traverses the graph to find connected concepts before answering
res_standard = client.queries.execute(
    query=question,
    kg_names=["KG_Companion_Demo"],
    include_citations=True
)
print(res_standard.answer)

# 2. Vector Bypass (Fast)
# Just finds similar text chunks. Good for simple lookups.
res_fast = client.queries.vector_bypass(
    query=question,
    kg_names=["KG_Companion_Demo"],
    top_k=3
)

Key Takeaways

Default to execute() for highest quality answers using the KG-RAG engine.
Use execute_unified() when you want to look at both "Facts" and "Episodes" simultaneously.
Use rerank() (not shown) if you just want to sort a list of documents by relevance.