# Search Model

# Search Model

Curiosity Workspace search is designed for structured + unstructured data, and integrates tightly with the graph model.

At a high level:

• Text search retrieves results based on keyword matching and ranking.
• Vector search retrieves results based on semantic similarity using embeddings.
• Hybrid search combines both to improve recall and precision.
• Facets / filters constrain results using properties and graph relationships.

# What gets indexed

Search indexing is explicit: you choose which node types and which fields are searchable.

Typical choices:

• titles and summaries
• identifiers (ticket IDs, serial numbers)
• descriptions and conversation text
• selected structured fields for faceting

# Text search

Text search is strong for:

• exact terms and identifiers
• short fields (names, titles)
• deterministic matching for compliance or audit use cases

Ranking can usually be tuned via:

• field-level boosting (e.g., “title counts more than body”)
• recency or timestamp-aware sort modes
• filtering and scoping (types, sources, permissions)

See Search → Text Search.

# Vector search (AI search)

Vector search is strong for:

• semantic similarity (“same issue, different words”)
• long text fields where users don’t know the right keywords
• multilingual or paraphrased content (depending on embedding model choice)

Key concepts:

• Embedding model turns text into vectors.
• Vector index stores vectors for fast similarity retrieval.
• Chunking may be required for long text fields to avoid losing context.

See Search → Vector Search.

# Filters and facets (including graph facets)

Search is frequently scoped by:

• property facets (e.g., Status=Open, Priority=P1)
• related facets (e.g., Ticket related to Customer=Acme via graph edges)
• time filters (useful for event-like nodes)

Graph-integrated faceting is often the differentiator: it lets you constrain search based on relationships rather than only local fields.

# Access control (conceptual)

In production environments, search must respect:

• user permissions
• document/node visibility
• tenant isolation (if multi-tenant)

See Administration → Permissions.

# Next steps

• How to configure text retrieval: Search → Text Search
• How to add embeddings: NLP → Embeddings