Operator Reference

This page is the compact map of ParadeDB’s SQL search surface. Use operators for common field predicates, and use @@@ with a pdb.* query builder when the query needs explicit options, parser syntax, or a query object. For example, these two queries express the same match-disjunction search:

SELECT id, description
FROM mock_items
WHERE description ||| 'running shoes';

SELECT id, description
FROM mock_items
WHERE description @@@ pdb.match_disjunction('running shoes');

Search Operators

Operator	Typical use	Function equivalent	Detailed reference
`@@@`	General ParadeDB predicate	`pdb.parse(...)`, `pdb.all()`, or builder	Query Parser, All
`\|\|\|`	Tokenized text match, any token	`pdb.match_disjunction(...)`	Match Disjunction
`&&&`	Tokenized text match, all tokens	`pdb.match_conjunction(...)`	Match Conjunction
`===`	Exact token match or token set	`pdb.term(...)`, `pdb.term_set(...)`	Term
`###`	Phrase match	`pdb.phrase(...)` or `pdb.phrase_array(...)`	Phrase
`##`	Proximity clause, any order	`pdb.proximity(...)`	Proximity
`##>`	Proximity clause, ordered	`pdb.proximity_in_order(...)`	Proximity

## and ##> build proximity clauses and are normally wrapped by @@@:

SELECT id, description
FROM mock_items
WHERE description @@@ ('sleek' ## 1 ## 'shoes');

Vector Operators

Vector search uses pgvector distance operators in ORDER BY ... LIMIT queries. ParadeDB can accelerate these when the vector column is in the ParadeDB index and the query also contains a ParadeDB predicate such as id @@@ pdb.all().

Operator	Distance	ParadeDB vector operator class
`<->`	L2	`vector_l2_ops`
`<=>`	Cosine	`vector_cosine_ops`
`<#>`	Inner product	`vector_ip_ops`

See Querying Vectors for examples and Tuning Recall and Latency for recall controls.

Operator Inputs

Operator	Right-hand side inputs	Notes
`@@@`	`text`, `pdb.query`, `pdb.proximityclause`, `searchqueryinput`	`text` is parsed with ParadeDB’s query parser. Query builders return `pdb.query`.
`\|\|\|`	`text`, `text[]`, `pdb.query`, `pdb.fuzzy`, `pdb.boost`	`text` is tokenized with the indexed field’s tokenizer.
`&&&`	`text`, `text[]`, `pdb.query`, `pdb.fuzzy`, `pdb.boost`	Same tokenization as match disjunction, but all query tokens must match.
`===`	`text`, `text[]`, `pdb.query`, `pdb.fuzzy`, `pdb.boost`	`text` is treated as a finalized token; `text[]` becomes a term set.
`###`	`text`, `text[]`, `pdb.query`, `pdb.slop`, `pdb.boost`	Matches ordered adjacent tokens unless slop is supplied.
`##`, `##>`	proximity clauses, text, or text arrays	Text and `text[]` are implicitly converted into proximity clauses.

Function Modifiers

Some operator inputs are modifier types rather than standalone predicates:

Modifier	Used with	Reference
`::pdb.fuzzy`	match, term	Fuzzy
`::pdb.boost`	match, term, phrase, builders	Relevance Tuning
`::pdb.const`	`@@@` and query builders	Relevance Tuning
`::pdb.slop`	`###`	Phrase

Choosing Operators or Functions

Use the symbolic operators for short, field-local predicates:

WHERE description &&& 'running shoes'
WHERE category === 'footwear'

Use @@@ with a pdb.* function when the query is not just one simple operator:

WHERE id @@@ pdb.parse('description:(running shoes) AND rating:>3')
WHERE description @@@ pdb.regex('key.*')
WHERE id @@@ pdb.all()

Use ordinary Postgres predicates for scalar filters. ParadeDB can push many of these predicates into the index when the filtered columns are indexed:

WHERE description ||| 'shoes' AND rating >= 4 AND in_stock IS TRUE

See Filtering for supported filter pushdown behavior.

Indexing

Full-Text / Lexical Search

Vector / Semantic Search

Hybrid Search

Aggregations

JOINs

Cheatsheets

Operator Reference

Search Operators

Vector Operators

Operator Inputs

Function Modifiers

Choosing Operators or Functions

​Search Operators

​Vector Operators

​Operator Inputs

​Function Modifiers

​Choosing Operators or Functions

Search Operators

Vector Operators

Operator Inputs

Function Modifiers

Choosing Operators or Functions