Curiosity
Three-column comparison of search layers: Text, Vector, and Hybrid with icons and descriptions.

The search layer

Three retrieval modes, always permission-aware. Results are filtered by the calling user's access before ranking — there is no post-processing step.


Mode How it works Best for
Text BM25 keyword ranking Identifiers, titles, exact terms
Vector Cosine similarity over embeddings Paraphrased questions, long text
Hybrid Text + vector, blended score Most enterprise data (default)

You choose what gets indexed. Not every field needs to be in search — explicit indexing keeps the index lean and ranking meaningful.

Facets let users narrow results by:

  • Node properties (status, priority, category)
  • Graph relationships (filter by customer, product, team)
  • Time (last 7 days, last quarter)

Search model