vector_search
Search Neo4j vector indexes using text queries to find similar nodes with sanitized properties and similarity scores for GraphRAG applications.
Instructions
Performs vector similarity search on a Neo4j vector index.
This tool embeds your text query using OpenAI and searches the specified vector index. Returns node IDs, labels, node properties (automatically sanitized), and similarity scores.
Automatic Sanitization (always applied):
Embedding property used by the vector index → automatically excluded (vector_search only)
Large lists (≥128 items) → replaced with placeholders
Large strings (≥10K chars) → truncated with suffix
Total response limited to 8000 tokens (results dropped if needed)
Property Selection:
Default (no return_properties): Returns ALL properties (sanitized)
With return_properties: Returns ONLY specified properties
Example: return_properties="pageNumber,id" → returns only these two
Check get_neo4j_schema_and_indexes for property warnings to avoid large fields
Performance Optimization: Internally fetches max(top_k × 2, 100) results to avoid local maximum problems in kANN algorithms.
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| text_query | Yes | The text query to search for. This will be embedded and used for similarity search. | |
| vector_index | Yes | The name of the vector index to search in. Use get_neo4j_schema_and_indexes to see available indexes. | |
| top_k | No | The number of most similar results to return. | |
| return_properties | No | Optional: Comma-separated list of properties to return (e.g., "pageNumber,id"). If not specified, returns all properties with automatic sanitization (large values are truncated). |