Skip to main content
Guide
Vector indexes rank node or edge embeddings by distance. Every definition requires a non-zero dimension and a distance metric.

Supported metrics

  • Cosine
  • Euclidean
  • Manhattan
The indexed value and every query vector must have exactly the declared dimension.

Create a vector index

Pass a tenant property as the final argument to partition the index. Start with the SDK’s vector-search operation for label Doc, property embedding, a query vector, and a result limit of 10. $distance is available on the hit stream; project it before traversing away from the hit if it must remain in the response.

Result limits

The server caps unrestricted vector search at 800 effective results per request. A narrower access bound from the surrounding plan can reduce this value. Traversal-scoped vector search checks the effective count after candidate intersection. It rejects the request when min(k, unique candidates) is greater than 800; it does not silently reduce k. A request with k above 800 can succeed when the candidate stream contains 800 or fewer unique entities. The candidate stream cannot contain more than 1,000,000 unique entities.

Operational notes

  • Creation returns before the backfill necessarily finishes.
  • Malformed source vectors can block the operation.
  • A new generation remains hidden until validation and activation succeed.
  • Dropping an index is also a durable lifecycle operation.

Next steps

Project search results

Preserve ranked hit metadata before continuing a traversal.

Vector prefiltering

Rank only an exact traversal-defined candidate set.

Troubleshoot index operations

Resolve blocked builds and lifecycle errors.