Smart Search

Ask in natural language. Inspect the query that actually ran.

Smart Search uses the project schema to translate a natural-language request into SearchQuery, executes that query, and returns both the records and the generated query. It is a query-generation surface, not a synonym for vector similarity.

Start free View docs

Natural-language querying is unsafe when the generated operation stays hidden.

An AI-generated answer is difficult to trust when software cannot inspect the labels, filters, relationship traversal, and pagination behind it. Smart Search keeps the generated SearchQuery available beside the result.

Before

Copy schema into a prompt by hand
Ask a model to invent database syntax
Execute opaque query text
Lose warnings and refinement context

With RushDB

Ground generation in the project schema
Receive ordinary RushDB SearchQuery
Inspect warnings before using the result
Pass currentQuery when refining a query-builder session

What it enables

Natural language with an inspectable execution boundary.

The result exposes data, total, searchQuery, and warnings. Applications can render the generated filters, log them for review, or reject query shapes outside an approved server-owned policy.

Schema-grounded generation

RushDB uses the labels, properties, and relationship structure available in the project instead of relying on a generic prompt alone.

Visible SearchQuery

The generated SearchQuery is attached to the SDK result so callers can explain which labels and filters produced the records.

Refinable context

Pass currentQuery when a user refines an existing dashboard or query-builder state rather than starting from an empty request.

How it works

Start with the smallest useful path.

Ask for a concrete result

Name the entity, conditions, and desired ordering or grouping as clearly as you would describe the query to another developer.

Inspect the generated query

Read result.searchQuery and result.warnings before exposing an AI-generated operation in a privileged or high-impact workflow.

Refine or constrain

Use currentQuery for conversational refinement and keep authorization, tenant scope, and allowed query shapes enforced in application code.

Flow

Prompt to inspectable result

Smart Search discovers the project structure, generates SearchQuery, executes it, and returns the query metadata beside the matching records.

Natural-language prompt

Project schema

Generated SearchQuery

Execution

Records + warnings

Implementation sketch

Return the generated query with the records.

The TypeScript and Python SDKs generate and execute the query in one call. The REST surface exposes generation separately, allowing the returned SearchQuery to be reviewed before execution.

from rushdb import RushDB

db = RushDB('RUSHDB_API_KEY')

results = db.ai.search(
    'Show open checkout incidents for enterprise accounts'
)

print(results.search_query)
print(results.warnings)
print(results.total)
print(results.data)

Know the operational boundary.

Smart Search is not vector search

Use records.vectorSearch or records.vector_search when you want direct similarity ranking over one indexed property. Smart Search generates an ordinary SearchQuery from natural language.

Read the SearchQuery guide

Generated does not mean automatically authorized

Inspect or validate generated labels, filters, and limits in server-owned code before using Smart Search for sensitive data or mutations.

Read the search documentation

Related guides

Go deeper on the concept.

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

Guide

Ontology-Aware Querying

Let agents query using the real labels, properties, and relationship paths in your data, discovered from a live schema instead of a hand-maintained ontology doc.

Read guide

Guide

Semantic + Relational Retrieval

Combine meaning-based semantic search with exact relational filters and graph relationships in one query, instead of stitching together separate systems.