FalkorDB Vector Search Fails Without Correct Data Type and Index Setup

Developers using FalkorDB for semantic search or GraphRAG often encounter errors, empty results, or slow full scans when calling db.idx.vector.queryNodes, even when data appears correctly stored. The root cause typically comes down to two mandatory conditions that must both be satisfied simultaneously for native vector search to work. First, embedding data must be stored using the vecf32() function to ensure it is saved as a native vector type, not a plain list or string. Second, a vector index must be explicitly created on the corresponding property, as the index and the correct data type together enable efficient Approximate Nearest Neighbor search. A common mistake is passing a Python float list directly through Cypher without wrapping it in vecf32(), which causes FalkorDB to store it as a List type that the vector index does not recognize.