Full-Text + JSON Search¶

Uni supports full-text search over string properties and JSON documents, plus JSON path predicates for nested fields.

What It Provides¶

Full-text indexes for string properties.
JSON full-text search with path targeting.
CONTAINS predicates on JSON documents.
CALL uni.fts.query(...) for BM25-scored search with relevance ranking.

Example¶

RustPython

use uni_db::Uni;

# async fn demo() -> Result<(), uni_db::UniError> {
let db = Uni::open("./my_db").build().await?;
let session = db.session();

let tx = session.tx().await?;
tx.execute("CREATE FULLTEXT INDEX doc_fts FOR (d:Doc) ON (d.title, d.body)")
    .await?;
tx.commit().await?;

let rows = session.query(
    "MATCH (d:Doc) WHERE d.body CONTAINS 'vector' RETURN d.title"
).await?;

println!("{:?}", rows);
# Ok(())
# }

import uni_db

db = uni_db.Uni.open("./my_db")
session = db.session()

tx = session.tx()
tx.execute("CREATE FULLTEXT INDEX doc_fts FOR (d:Doc) ON (d.title, d.body)")
tx.commit()

rows = session.query(
    "MATCH (d:Doc) WHERE d.body CONTAINS 'vector' RETURN d.title"
)
print(rows)

FTS Query Procedure¶

For BM25-scored search with normalized relevance scores:

-- Create a fulltext index
CREATE FULLTEXT INDEX article_content FOR (a:Article) ON (a.content)

-- Query with relevance scoring
CALL uni.fts.query('Article', 'content', 'database optimization', 20)
YIELD node, score
RETURN node.title, score
ORDER BY score DESC

Parameters:

label - Node label to search
property - Text property with inverted index
search_term - Search query string
k - Number of results
threshold (optional) - Minimum score (0-1)

Yields:

vid - Vertex ID
score - Normalized BM25 score (0-1)
node - Full node with properties

How It Works¶

Auto-Build on Index Creation¶

When you run CREATE FULLTEXT INDEX, Uni automatically builds the FTS index over existing data. There is no need to call rebuild_indexes() manually — the index is ready to query immediately after creation.

-- This both creates AND builds the index in one step
CREATE FULLTEXT INDEX article_content FOR (a:Article) ON (a.content)

Immediate Write Visibility (L0 Buffer)¶

FTS queries see unflushed writes from the in-memory L0 buffer. This means data is searchable immediately after a write, without waiting for a flush to persistent storage. The query engine merges L0 buffer results with on-disk index results transparently.

-- Write and search in the same session — no flush needed
CREATE (a:Article {title: 'New Article', content: 'graph database optimization'})

-- This will find the article even before flush
CALL uni.fts.query('Article', 'content', 'optimization', 10)
YIELD node, score
RETURN node.title, score

Use Cases¶

Search across documents, notes, or product descriptions.
JSON documents with nested fields.
Hybrid filters that combine text search with graph structure.

When To Use¶

Use full-text or JSON search when keyword matching and relevance ranking matter more than exact equality on fields.