Vector + graph search

Retrieve meaning and connected context from one backend.

Semantic search is useful when the query is fuzzy. Graph traversal and structured filters add the context that similarity alone cannot provide.

Start free View docs

The nearest text match is not always the complete answer.

A vector store can find a related chunk while missing the product, customer, policy, or decision linked to it. Application-side joins add latency and another failure mode.

Before

Search vectors in one store
Load business records from another database
Join results in application code
Debug stale sync state when context disagrees

With RushDB

Index the text property you want ranked
Prefilter semantic search with exact fields
Traverse explicit relationships through SearchQuery
Keep records, vectors, and graph context together

What it enables

Retrieve the answer, not just the nearest chunk.

RushDB semantic search narrows candidates with labels and where filters before ranking, then your workflow can follow relationships when the answer depends on connected records.

Ground fuzzy search with facts

Use exact filters for tenant, status, category, permission scope, or workflow state before similarity ranking happens.

Keep evidence connected

Documents, entities, policies, decisions, and users can stay in one graph instead of being rejoined after vector search.

Support RAG and apps

The same backend can power agent recall, knowledge-base retrieval, faceted search, and operational dashboards.

How it works

Start with the smallest useful path.

Rank by meaning

Use ai.search for natural-language retrieval against an indexed string property.

Narrow with exact filters

Apply a where clause before scoring when the result must stay within a tenant, status, category, or workflow.

Follow connected records

Use SearchQuery relationship traversal when the answer needs related entities instead of only the nearest text.

Flow

Meaning plus structure

Start with a semantic query when language is fuzzy, add exact filters to keep scope safe, then use graph context when the product needs connected evidence.

Natural-language query

Exact filters

Semantic ranking

Connected records

Grounded result

Implementation sketch

Prefilter semantic recall with ordinary fields.

This keeps similarity results inside one agent scope. Use SearchQuery traversal when the workflow also needs connected records.

Know the operational boundary.

Relationships still need a source

Nested imports preserve parent-child edges. Domain relationships come from approved suggested patterns or explicit application writes.

Review suggested relationship patterns

Use the simple path first

Start with ai.search plus where filters. Move into SearchQuery vector expressions when you need aggregation or multi-hop retrieval in the same query.

Read the semantic search guide

Related guides

Go deeper on the concept.

These guides explain the product category and implementation tradeoffs behind this feature.

Guide

Vector Database vs Memory Layer

Compare vector databases and memory layers for AI agents. Learn when semantic search is enough and when records, relationships, and ontology are needed.

Read guide

Guide

Knowledge Graph Memory

Learn how graph memory keeps documents, entities, citations, users, and decisions connected for AI agents and GraphRAG workflows.