What problem does it solve? Setting up vector search in Neo4j involves version-specific syntax, index configuration choices, and embedding pipeline pitfalls that silently produce wrong results. This Skill provides the correct Cypher patterns for creating vector indexes, ingesting embeddings, and running similarity search across Neo4j versions. ## Core Features & Use Cases - Vector Index Management: Create node or relationship vector indexes with dimensions, similarity functions, HNSW parameters, and quantization options, then poll until ONLINE. - Similarity Search: Run nearest-neighbor queries using the SEARCH clause (Neo4j 2026.01+) or the db.index.vector.queryNodes() procedure fallback, with in-index filtering on declared properties. - Embedding Ingestion: Batch-store embeddings with Python UNWIND patterns or generate them in-Cypher with ai.text.embed(), plus hybrid search combining vector and fulltext sources via WRRF ranking. - Use Case: You are building a document Q&A system on Neo4j and need to index text chunk embeddings, then retrieve the most relevant chunks filtered by source and language before passing them to an LLM. ## Quick Start Ask the assistant to create a Neo4j vector index on Chunk nodes with 1536 dimensions and cosine similarity, then run a semantic search query against it.
