Query - TopK

TopK provides a data frame-like syntax for querying documents. It features built-in semantic search, text search, vector search, and metadata filtering capabilities. With TopK’s declarative query builder, you can easily select fields, chain filters, and apply vector/text search in a composable manner.

Query structure

In TopK, a query consists of multiple stages:

Select stage

Select static or computed fields that will be returned in the query results. These fields can be used in stages such as Filter or TopK.

Filter stage

Filter the documents that will be returned in the query results. Filters can be applied to static fields, computed fields such as vector_distance() or semantic_similarity(), or custom properties computed inside select().

Sort stage

Order results by an expression (ascending or descending).

Limit stage

Return at most k results.

Count stage

Return the total number of documents matching the query.

All queries must have either Sort + Limit or Count collection stage.

You can stack multiple select and filter stages in a single query.

A typical query in TopK looks as follows:

Select

The select() function is used to initialize the select stage of a query. It accepts a key-value pair of field names and field expressions:

from topk_sdk.query import select, field

client.collection("books").query(
  select(
    "published_year", # elect the static fields directly
    title=field("title"),
  )
  ...
)

import { select, field } from "topk-js/query";

await client.collection("books").query(
  select({
    title: field("title"),
  })
  ...
)

Select expressions

Use a field() function to select fields from a document. In the select stage, you can also rename existing fields or define computed fields using function expressions.

from topk_sdk.query import select, field

docs = client.collection("books").query(
    select(
        "title", # the actual "title" field from the document
        year=field("published_year"), # renamed field
        year_plus_ten=field("published_year") + 10, # computed field
    )
)

import { select, field } from "topk-js/query";

const docs = await client.collection("books").query(
  select({
    title: field("title"), // the actual "title" field from the document
    year: field("published_year"), // renamed field
    year_plus_ten: field("published_year").add(10), // computed field
  })
);

SELECT
  title,                              -- static field
  published_year AS year,             -- renamed field
  published_year + 10 AS year_plus_ten  -- computed field
FROM books
LIMIT 10;

Function expressions

Function expressions are used to define computed fields that will be included in your query results. TopK currently supports four main function expressions:

vector_distance(field, vector): Computes distance between vectors for vector search. This function is available for all dense and sparse vector types.
bm25_score(): Calculates relevance scores using the BM25 algorithm for keyword search
semantic_similarity(field, query): Measures semantic similarity between the provided text query and the field’s embedding
multi_vector_distance(field, matrix): Computes MaxSim distance for multi-vector (matrix) fields. Requires a multi_vector_index() on the field. See multi-vector search.

Vector distance

The vector_distance() function is used to compute the vector score between a query vector and a vector field in a collection. There are multiple ways to represent a query vector:

Dense vectors:
- [0.1, 0.2, 0.3, ...] - Array of numbers resolved as a dense float32 vector
- f32_vector([...]) - Helper function returning a dense float32 vector
- f16_vector([...]) - Helper function returning a dense float16 vector
- f8_vector([...]) - Helper function returning a dense float8 vector
- u8_vector([...]) - Helper function returning a dense u8 vector
- i8_vector([...]) - Helper function returning a dense i8 vector
- binary_vector([...]) - Helper function returning a binary vector
Sparse vectors:
- { 0: 0.1, 1: 0.2, 2: 0.3, ... } - Mapping from index → value resolved as a sparse float32 vector
- f32_sparse_vector({ ... }) - Helper function returning a sparse float32 vector
- u8_sparse_vector({ ... }) - Helper function returning a sparse u8 vector

Optionally, uses can provide skip_refine=True to bypass the internal distance refinement step. This will improve performance for queries with large top_k at the cost of lower accuracy.

We don’t recommend using skip_refine=True unless you’re using a large top_k.

To use the vector_distance() function, you must have a vector index defined on the field you’re computing the vector distance against:

from topk_sdk.query import select, field, fn

docs = client.collection("books").query(
    select(
        "title",
        title_similarity=fn.vector_distance(
          "title_embedding",
          [0.1, 0.2, 0.3, ...] # embedding for "animal"
        )
    )
    .sort(field("title_similarity"), asc=False).limit(10)
)

# Example result:
[
  {
    "_id": "2",
    "title": "To Kill a Mockingbird",
    "title_similarity": 0.7484796643257141
  },
  {
    "_id": "1",
    "title": "The Catcher in the Rye",
    "title_similarity": 0.5471329569816589
  }
]

import { select, field, fn } from "topk-js/query";

const docs = await client.collection("books").query(
  select({
    title: field("title"),
    title_similarity: fn.vectorDistance(
      "title_embedding",
      [0.1, 0.2, 0.3 /* embedding for "animal" */]
    ),
  })
);

// Example result:
[
  {
    _id: '2',
    title: 'To Kill a Mockingbird',
    title_similarity: 0.7484796643257141
  },
  {
    _id: '1',
    title: 'The Catcher in the Rye',
    title_similarity: 0.5471329569816589,
  }
]

SELECT
  title,
  vector_distance(title_embedding, '[0.1, 0.2, ..., 0.3]'::f32_vector) AS title_similarity
FROM books
ORDER BY title_similarity DESC
LIMIT 10;

BM25 Score

The BM25 score is a relevance score that can be used to score documents based on their text content. To use the fn.bm25_score() in your query, you must include a match predicate in your filter stage.

To use the fn.bm25_score() function, you must have a keyword index defined in your collection schema.

from topk_sdk.query import select, field, fn, match

docs = client.collection("books").query(
    select(
        "title",
        text_score=fn.bm25_score(),
    )
    .filter(match("Good")) # must include a match predicate
    .sort(field("text_score"), asc=False).limit(10)
)

# Example result:
[
  {
    "_id": "1",
    "title": "Good Night, Bat! Good Morning, Squirrel!",
    "text_score": 0.2447269707918167
  },
  {
    "_id": "2",
    "title": "Good Girl, Bad Blood",
    "text_score": 0.20035339891910553
  }
]

import { select, field, fn, match } from "topk-js/query";

const results = await client.collection("books").query(
  select({
    title: field("title"),
    text_score: fn.bm25Score(),
  })
    .filter(match("Good"))
    .sort(field("text_score"), false).limit(10)
);

// Example result:
[
  {
    _id: '1',
    title: 'Good Night, Bat! Good Morning, Squirrel!',
    text_score: 0.2447269707918167,
  },
  {
    _id: '2',
    title: 'Good Girl, Bad Blood',
    text_score: 0.20035339891910553
  }
]

SELECT
  title,
  bm25_score() AS text_score
FROM books
WHERE match_any(title, 'Good')
ORDER BY text_score DESC
LIMIT 10;

Semantic similarity

The semantic_similarity() function is used to compute the similarity between a text query and a text field in a collection. To use the semantic_similarity() function, you must have a semantic index defined on the field you’re computing the similarity on.

from topk_sdk.query import select, field, fn

docs = client.collection("books").query(
    select(
        "title",
        title_similarity=fn.semantic_similarity("title", "animal"),
    )
    .sort(field("title_similarity"), asc=False).limit(10)
)

# Example result:
[
  {
    "_id": "2",
    "title": "To Kill a Mockingbird",
    "title_similarity": 0.7484796643257141
  },
  {
    "_id": "1",
    "title": "The Catcher in the Rye",
    "title_similarity": 0.5471329569816589
  }
]

import { select, field, fn } from "topk-js/query";

const docs = await client.collection("books").query(
  select({
    title: field("title"),
    title_similarity: fn.semanticSimilarity("title", "animal"),
  })
    .sort(field("title_similarity"), false).limit(10)
);

// Example result:
[
  {
    _id: '2',
    title: 'To Kill a Mockingbird',
    title_similarity: 0.7484796643257141
  },
  {
    _id: '1',
    title: 'The Catcher in the Rye',
    title_similarity: 0.5471329569816589
  }
]

SELECT
  title,
  semantic_similarity(title, 'animal') AS title_similarity
FROM books
ORDER BY title_similarity DESC
LIMIT 10;

Multi-vector distance

The multi_vector_distance() function computes the MaxSim score between a query matrix and a matrix field in a collection. Use it for multi-vector (late-interaction) retrieval when documents are stored as N x D embedding matrices. To use multi_vector_distance(), you must have a multi_vector_index() defined on the field. The query matrix can be a list of lists (defaults to f32), a numpy array (type inferred from dtype), or a matrix() instance. The optional candidates parameter limits the number of candidate vectors considered during search.

from topk_sdk.query import select, field, fn

docs = client.collection("passages").query(
    select(
        "content",
        dist=fn.multi_vector_distance(
            "token_embeddings",
            [[0.1, 0.2, ...], [0.4, 0.5, ...]],  # query matrix
            candidates=100,
        ),
    )
    .sort(field("dist"), asc=False).limit(10)
)

import { select, field, fn } from "topk-js/query";

const docs = await client.collection("passages").query(
  select({
    content: field("content"),
    dist: fn.multiVectorDistance(
      "token_embeddings",
      [[0.1, 0.2, ...], [0.4, 0.5, ...]],  // query matrix
      100  // optional: candidates
    ),
  })
    .sort(field("dist"), false).limit(10)
);

SELECT
  content,
  multi_vector_distance(
    token_embeddings,
    '[[0.1, 0.2, ..., 0.0], [0.4, 0.5, ..., 0.1]]'::f32_matrix,
    100
  ) AS dist
FROM passages
ORDER BY dist DESC
LIMIT 10;

See multi-vector search for schema setup and ingestion details.

Advanced select expressions

TopK doesn’t only let you select static fields from your documents or computed fields using function expressions. You can also use TopK powerful expression language to select fields by chaining arbitrary logical expressions:

from topk_sdk.query import select, field

select(
  weight_in_grams=field("weight").mul(1000),
  is_adult=field("age").gt(18),
  published_in_nineteenth_century=field("published_year") >= 1800 & field("published_year") < 1900,
)

import { select, field } from "topk-js/query";

select({
  weight_in_grams: field("weight").mul(1000),
  is_adult: field("age").gt(18),
  published_in_nineteenth_century: field("published_year")
    .gte(1800)
    .and(field("published_year").lt(1900)),
});

Filtering

You can filter documents by metadata, keywords, custom properties computed inside select() (e.g. vector similarity or BM25 score) and more. Filter expressions support all

logical operators,
comparison operators,
mathematical operators.

Metadata filtering

.filter(
    field("published_year") > 1980
)

.filter(field("published_year").gt(1980))

WHERE published_year > 1980

Keyword search

The match() function is the backbone of keyword search in TopK. It allows you to search for documents that contain specific keywords or phrases. You can configure the match() function to:

Match on multiple terms
Match only on specific fields
Use weights to prioritize certain terms

The match() function accepts the following parameters:

token

string

required

String token to match. Can also contain multiple terms separated by a delimiter which is any non-alphanumeric character.

options.field

string

Field to match on. If not provided, the function will match on all fields.

options.weight

number

Weight to use for matching. If not provided, the function will use the default weight(1.0).

options.all

boolean

Use all parameter when a text must contain all terms(separated by a delimeter)

when all is false (default) it’s an equivalent of OR operator
when all is true it’s an equivalent of AND operator

Searching for a term like "catcher" in your documents is as simple as using the match() function in the filter stage of your query:

from topk_sdk.query import match

.filter(
    match("catcher")
)

import { match } from "topk-js/query";

.filter(match("catcher"))

WHERE match_any(title, 'catcher')

Match multiple terms

The match() function can be configured to match all terms when using a delimiter. A term delimiter is any non-alphanumeric character. To ensure that all terms are matched, use the all parameter:

from topk_sdk.query import match

.filter(match("catcher|rye", field="title", all=True))

import { match } from "topk-js/query";

.filter(match("catcher|rye", { field: "title", weight: 1, all: true }))

WHERE match_all(title, 'catcher rye')

Give weight to specific terms

You can give weight to specific terms by using the weight parameter:

from topk_sdk.query import match

.filter(match("catcher", weight=2.0) | match("rye", weight=1.0))

import { match } from "topk-js/query";

.filter(match("catcher", { weight: 2.0 }).or(match("rye", { weight: 1.0 })))

Combine keyword search and metadata filtering

You can combine metadata filtering and keyword search in a single query by stacking multiple filter stages. In the example below, we’re searching for documents that contain the keyword "catcher" and were published in 1997, or between 1920 and 1980.

.filter(
    match("catcher")
)
.filter(
    field("published_year") == 1997 | (field("published_year") >= 1920 & field("published_year") <= 1980)
)

.filter(
  match("catcher")
)
.filter(
  field("published_year").eq(1997).or(field("published_year").gte(1920).and(field("published_year").lte(1980)))
)

WHERE match_any(title, 'catcher')
  AND (published_year = 1997
    OR (published_year >= 1920 AND published_year <= 1980))

Operators

When writing queries, you can use the following operators for:

field selection
filtering
topk collection

Logical operators

Logical operators combine multiple expressions by applying boolean logic and conditions.

and

The and operator can be used to combine multiple logical expressions.

.filter(
    field("published_year") == 1997 & field("title") == "The Catcher in the Rye"
)

# or

.filter(
    field("published_year").eq(1997).and_(field("title").eq("The Catcher in the Rye"))
)

.filter(
  field("published_year").eq(1997).and(field("title").eq("The Catcher in the Rye"))
)

or

The or operator can be used to combine multiple logical expressions.

.filter(
    field("published_year") == 1997 | field("title") == "The Catcher in the Rye"
)

# or

.filter(
    field("published_year").eq(1997).or_(field("title").eq("The Catcher in the Rye"))
)

.filter(
  field("published_year").eq(1997).or(field("title").eq("The Catcher in the Rye"))
)

not

The not helper can be used to negate a logical expression. It takes an expression as an argument and inverts its logic.

from topk_sdk.query import field, not_

.filter(
    not_(field("title").contains("Catcher"))
)

import { field, not } from "topk-js/query";

.filter(
    not(field("title").contains("Catcher"))
)

all

The all() helper evaluates to true if each expression in the array is true. It’s equivalent to applying the logical AND operator across all expressions.

from topk_sdk.query import all, field

.filter(
    all([
        field("published_year") >= 1900,
        field("published_year") <= 2000,
        field("title").is_not_null()
    ])
)

import { all, field } from "topk-js/query";

.filter(
  all([
    field("published_year").gte(1900),
    field("published_year").lte(2000),
    field("title").isNotNull()
  ])
)

This is equivalent to:

field("published_year") >= 1900 & field("published_year") <= 2000 & field("title").is_not_null()

field("published_year").gte(1900).and(field("published_year").lte(2000)).and(field("title").isNotNull())

any

The any() helper evaluates to true if at least one expression in the array is true. It’s equivalent to applying the logical OR operator across all expressions.

from topk_sdk.query import any, field

.filter(
    any([
        field("genre") == "fiction",
        field("genre") == "mystery",
        field("genre") == "thriller"
    ])
)

import { any, field } from "topk-js/query";

.filter(
  any([
    field("genre").eq("fiction"),
    field("genre").eq("mystery"),
    field("genre").eq("thriller")
  ])
)

This is equivalent to:

field("genre") == "fiction" | field("genre") == "mystery" | field("genre") == "thriller"

field("genre").eq("fiction").or(field("genre").eq("mystery")).or(field("genre").eq("thriller"))

choose

The choose operator evaluates a condition and returns the first argument if the condition is true, else the second argument.

select(
  summary=(field("book_type") == "fiction").choose(
      field("plot_summary"),
      field("technical_summary")
  )
)

select({
  summary: field("book_type").eq("fiction").choose(
    field("plot_summary"),
    field("technical_summary")
  ),
})

boost

The boost operator multiplies the scoring expression by the provided boost value if the condition is true. Otherwise, the scoring expression is unchanged (multiplied by 1).

select(
  summary_distance=fn.vector_distance("summary_embedding", [2.3] * 16)
).sort(
  field("summary_distance").boost(field("summary").match_all("deep learning"), 1.5),
  asc=False
).limit(10)
# this boost expression is equivalent to
# field("summary_distance") * (field("summary").match_all("deep learning").choose(1.5, 1.0)),

select({
  summary_distance: fn.vectorDistance("summary_embedding", Array(16).fill(2.3)),
}).sort(
  field("summary_distance").boost(field("summary").matchAll("deep learning"), 1.5),
  false
).limit(10)
// this boost expression is equivalent to
// field("summary_distance").mul(field("summary").matchAll("deep learning").choose(1.5, 1.0)),

coalesce

The coalesce operator replaces null values with a provided value.

select(importance=field("nullable_importance").coalesce(1.0))

select({
  importance: field("nullable_importance").coalesce(1.0)
})

Comparison operators

Comparison operators provide various logical, numerical and string functions that evaluate to true or false.

eq

The eq operator can be used to match documents that have a field with a specific value.

.filter(
    field("published_year") == 1997
)

# or

.filter(
    field("published_year").eq(1997)
)

.filter(field("published_year").eq(1997))

ne

The ne operator can be used to match documents that have a field with a value that is not equal to a specific value.

.filter(
    field("published_year") != 1997
)

# or

.filter(
    field("published_year").ne(1997)
)

.filter(field("published_year").ne(1997))

is_null

The is_null operator can be used to match documents that have a field with a value that is null.

.filter(
    field("title").is_null()
)

.filter(field("title").isNull())

is_not_null

The is_not_null operator can be used to match documents that have a field with a value that is not null.

.filter(
    field("title").is_not_null()
)

.filter(field("title").isNotNull())

gt

The gt operator can be used to match documents that have a field with a value greater than a specific value. For strings, it uses lexicographic order.

.filter(
    field("published_year") > 1997
)

# or

.filter(
    field("published_year").gt(1997)
)

.filter(field("published_year").gt(1997))

gte

The gte operator can be used to match documents that have a field with a value greater than or equal to a specific value. For strings, it uses lexicographic order.

.filter(
    field("published_year") >= 1997
)

# or

.filter(
    field("published_year").gte(1997)
)

.filter(field("published_year").gte(1997))

lt

The lt operator can be used to match documents that have a field with a value less than a specific value. For strings, it uses lexicographic order.

.filter(
    field("published_year") < 1997
)

# or

.filter(
    field("published_year").lt(1997)
)

.filter(field("published_year").lt(1997))

lte

The lte operator can be used to match documents that have a field with a value less than or equal to a specific value. For strings, it uses lexicographic order.

.filter(
    field("published_year") <= 1997
)

# or

.filter(
    field("published_year").lte(1997)
)

.filter(field("published_year").lte(1997))

starts_with

The starts_with operator can be used on string fields to match documents that start with a given prefix. This is especially useful in multi-tenant applications where document IDs can be structured as {tenant_id}/{document_id} and starts_with can then be used to scope the query to a specific tenant. Also supports list-of-string fields for prefix filtering on array elements (e.g. field("tags").starts_with("fiction")).

.filter(
    field("_id").starts_with("tenant_123/")
)

.filter(field("_id").startsWith("tenant_123/"))

contains

The contains operator can be used on both text fields and list fields to match documents that include a specific value. For text fields, it matches documents that include a specific substring (case-sensitive). For list fields, it matches documents where the field of type list contains the specified value.

Text fields: Matches documents that include a specific substring. It is case-sensitive and avoids the text processing pipeline (tokenization and stemming) used by the match() function. This makes it particularly useful when you need exact substring matching or want to provide your own pre-processed tokens. Unlike match(), the contains operator can be used without requiring a keyword index.
List fields: Matches documents where the list field contains the specified value. The value can be a literal or a field reference. You can also use a list of strings with a keyword index if you want to provide your own tokens instead of using the text processing pipeline.

# String contains
.filter(
    field("title").contains("Catcher")
)

# List contains
.filter(
    # Tags is a text list field: ["novel", "fiction", "science-fiction"]
    field("tags").contains("novel")
)

# List contains combined with a field reference
.filter(
    field("codes").contains(field("slug"))
)

// String contains
.filter(field("title").contains("Catcher"))

// List contains
.filter(field("tags").contains("novel"))

// List contains combined with a field reference
.filter(field("codes").contains(field("slug")))

The contains operator works exactly the same as the in operator, but with reversed operands: x CONTAINS y is equivalent to y IN x. Both operators are provided for convenience and to make queries more readable.

in

The in (or in_ in Python) operator checks if a field value is present in a list of values, string literal or another field. It can be used in several ways:

Field in list: Checks if a field value is present in a list of literal values.
Field in string: Checks if a string field is a substring of another string. Unlike the match(), this avoids the text processing pipeline (tokenization and stemming) and performs exact substring matching.
Field in field: Checks if a field value is present in another field.

# Field in list of literals
.filter(
    field("published_year").in_([1999, 1988, 1997])
)

# Field in string
.filter(
    field("slug").in_("harry-potter lord-of-the-rings")
)

# Field in another field (list)
.filter(
    field("_id").in_(field("tags"))
)

// Field in list of literals
.filter(field("published_year").in([1999, 1988, 1997]))

// Field in string
.filter(field("slug").in("harry-potter lord-of-the-rings"))

// Field in another field (list)
.filter(field("_id").in(field("tags")))

The in operator works exactly the same as the contains operator, but with reversed operands: y IN x is equivalent to x CONTAINS y. Both operators are provided for convenience and to make queries more readable.

match_all

The match_all operator returns true if all terms in the query are present in the field with a keyword index.

.filter(
  field("summary").match_all("love marriage england")
)
# you can also pass a list of strings:
.filter(
  field("summary").match_all(["love", "marriage", "england"])
)

.filter(field("summary").matchAll("love marriage england"))
// you can also pass an array of strings:
.filter(field("summary").matchAll(["love", "marriage", "england"]))

When using a match_all operator against a text field, it must be used in conjunction with a keyword index defined in your collection schema.

match_any

The match_any operator returns true if any term in the query is present in the field with a keyword index.

.filter(
    field("summary").match_any("love ring")
)
# you can also pass a list of strings:
.filter(
  field("summary").match_any(["love", "ring"])
)

.filter(field("summary").matchAny("love ring"))
// you can also pass an array of strings:
.filter(field("summary").matchAny(["love", "ring"]))

When using a match_any operator against a text field, it must be used in conjunction with a keyword index defined in your collection schema.

regexp_match

The regexp_match operator returns true if the field value matches the regular expression. Internally, this uses Rust’s regex crate to evaluate the regular expression.

.filter(
    field("summary").regexp_match("^love")
)
# Optionally, you can pass flags to the regular expression
.filter(
    field("summary").regexp_match("^love", "i")
)

.filter(field("summary").regexpMatch("^love"))
// Optionally, you can pass flags to the regular expression
.filter(field("summary").regexpMatch("^love", "i"))

Mathematical operators

Mathematical operators perform computations on numbers.

add

The add operator can be used to add two numbers.

.filter(
    field("published_year") + 1997
)

# or

.filter(
    field("published_year").add(1997)
)

.filter(field("published_year").add(1997))

sub

The sub operator can be used to subtract two numbers.

.filter(
    field("published_year") - 1997
)

# or

.filter(
    field("published_year").sub(1997)
)

.filter(field("published_year").sub(1997))

mul

The mul operator can be used to multiply two numbers.

.filter(
    field("published_year") * 1997
)

# or

.filter(
    field("published_year").mul(1997)
)

.filter(field("published_year").mul(1997))

div

The div operator can be used to divide two numbers.

.filter(
    field("published_year") / 1997
)

# or

.filter(
    field("published_year").div(1997)
)

.filter(field("published_year").div(1997))

abs

The abs operator returns the absolute value of a number, which is useful for calculating distances or differences.

from topk_sdk.query import abs

# Find books published closest to 1990
select(
    delta=abs(field("published_year").sub(1990))
)

import { abs } from "topk-js/query";

// Find books published closest to 1990
select({
  delta: abs(field("published_year").sub(1990))
})

min

The min operator returns the smaller of two values, commonly used for clamping or setting upper bounds. It can work with both scalar values and other fields or expressions. For strings, it uses lexicographic order.

from topk_sdk.query import min

# Clamp BM25 scores to a maximum of 2.0
select(
    clamped_score=min(field("bm25_score"), 2.0)
)

# Take the lower of critic score vs user rating
select(
    conservative_score=min(field("critic_score"), field("user_rating"))
)

import { min } from "topk-js/query";

// Clamp BM25 scores to a maximum of 2.0
select({
  clamped_score: min(field("bm25_score"), 2.0)
})

// Take the lower of critic score vs user rating
select({
  conservative_score: min(field("critic_score"), field("user_rating"))
})

max

The max operator returns the larger of two values, commonly used for clamping or setting lower bounds. It can work with both scalar values and other fields or expressions. For strings, it uses lexicographic order.

from topk_sdk.query import max

# Ensure minimum relevance score of 1.5
select(
    boosted_score=max(field("relevance_score"), 1.5)
)

# Take the higher of critic score vs user rating
select(
    best_score=max(field("critic_score"), field("user_rating"))
)

import { max } from "topk-js/query";

// Ensure minimum relevance score of 1.5
select({
  boosted_score: max(field("relevance_score"), 1.5)
})

// Take the higher of critic score vs user rating
select({
  best_score: max(field("critic_score"), field("user_rating"))
})

ln

The ln operator calculates the natural logarithm, useful for logarithmic scaling and dampening large values.

# Apply logarithmic dampening to scores
select(
    log_score=(field("raw_score") + 1).ln()
)

// Apply logarithmic dampening to scores
select({
  log_score: field("raw_score").add(1).ln()
})

exp

The exp operator calculates the exponential function (e^x), useful for exponential scaling and boosting.

# Apply exponential boosting to BM25 scores
select(
    boosted_score=(field("bm25_score") * 1.5).exp()
)

// Apply exponential boosting to BM25 scores
select({
  boosted_score: field("bm25_score").mul(1.5).exp()
})

sqrt

The sqrt operator calculates the square root, useful for dampening values and creating non-linear transformations.

# Dampen large distance values
select(
    dampened_distance=field("vector_distance").sqrt()
)

// Dampen large distance values
select({
  dampened_distance: field("vector_distance").sqrt()
})

square

The square operator multiplies a number by itself (x²), useful for amplifying differences and creating quadratic transformations.

# Create quadratic penalty for age differences
select(
    age_penalty=(field("user_age") - 50).square()
)

// Create quadratic penalty for age differences
select({
  age_penalty: field("user_age").sub(50).square()
})

Collection

All queries must have a collection stage. Currently, we only support topk() and count() collectors.

topk

Use the topk() function to return the top k results. The topk() function accepts the following parameters:

expr

LogicalExpression

required

The logical expression to sort the results by.

number

required

The number of results to return.

asc

boolean

required

Whether to sort the results in ascending order.

To get the top 10 results with the highest title_similarity, you can use the following query:

.sort(field("title_similarity"), asc=False).limit(10)

.sort(field("title_similarity"), false).limit(10)

ORDER BY title_similarity DESC
LIMIT 10

The topk() stage is equivalent to applying sort(expr, asc) followed by limit(k). It is a convenience shorthand for the common pattern of ordering results by a scoring expression and returning the top k.

limit

Use the .limit(k) stage to return at most k results. Can be used with or without topk().

client.collection("books").query(
    select("title", "author"),
    filter=field("year") >= 2000,
).limit(50)

await client.collection("books").query(
  select({ title: field("title"), author: field("author") }),
  { filter: field("year").gte(2000) }
).limit(50);

SELECT title, author
FROM books
WHERE published_year >= 2000
LIMIT 50;

sort

Use the .sort(expr, asc) stage to sort results by an expression. Use asc=True for ascending order, asc=False for descending order.

client.collection("books").query(
    select("title", "author", "year"),
    filter=field("author") == "George Orwell",
).sort(field("year"), asc=True)

await client.collection("books").query(
  select({ title: field("title"), author: field("author"), year: field("year") }),
  { filter: field("author").eq("George Orwell") }
).sort(field("year"), true);

SELECT title, author, published_year AS year
FROM books
WHERE author = 'George Orwell'
ORDER BY published_year ASC
LIMIT 100;

count

Use the count() function to get the total number of documents matching the query. If there are no filters then count() will return the total number of documents in the collection.

# Count the total number of documents in the collection
.count()

// Count the total number of documents in the collection
.count()

-- Count all documents
SELECT COUNT(*) FROM books;

-- Count with a filter
SELECT COUNT(*) FROM books WHERE published_year > 2000;

When writing queries, remember that they all require the topk or count function at the end.

Query options

The query() method accepts a consistency parameter to control read consistency: indexed, balanced (default), or strong. See the consistency concept for details.

LSN-based Consistency

TopK supports LSN (Log Sequence Number) based consistency for ensuring read-after-write consistency. When you perform a write operation (like upsert), you receive an LSN as a string that represents the sequence number of that write in the system’s log. You can use this LSN in subsequent queries to ensure that the query only returns results that are at least as recent as that write operation.

How it works

Write operation: When you call lsn = client.collection().upsert(), you receive an LSN
Query with LSN: Pass that LSN to client.collection().query(..., lsn=lsn)
Consistency guarantee: If the write is not yet available in the read path, the query will be rejected and the client will automatically retry

This approach ensures that your queries always see the results of your recent writes, providing strong consistency guarantees when needed.

# Upsert a document and get the LSN
lsn = client.collection("books").upsert([
    {"_id": "1984", "title": "1984", "author": "George Orwell", "year": 1949}
])

# Query with LSN to ensure consistency (optionally use consistency="strong")
results = client.collection("books").query(
    select("title", "author", "year")
    .filter(field("author") == "George Orwell")
    .sort(field("year"), asc=False).limit(10),
    lsn=lsn,
    consistency="strong"
)

// Upsert a document and get the LSN
const lsn = await client.collection("books").upsert([
  { _id: "1984", title: "1984", author: "George Orwell", "year": 1949 }
]);

// Query with LSN to ensure consistency (optionally use consistency: "strong")
const results = await client.collection("books").query(
  select({ title: field("title"), author: field("author"), year: field("year") })
  .filter(field("author").eq("George Orwell"))
  .sort(field("title"), false).limit(10),
  { lsn: lsn, consistency: "strong" }
)

Using LSN-based consistency may increase query latency as the system needs to verify that the specified LSN has been processed before returning results.

​Query structure

​Select

​Select expressions

​Function expressions

​Vector distance

​BM25 Score

​Semantic similarity

​Multi-vector distance

​Advanced select expressions

​Filtering

​Metadata filtering

​Keyword search

​Match multiple terms

​Give weight to specific terms

​Combine keyword search and metadata filtering

​Operators

​Logical operators

​and

​or

​not

​all

​any

​choose

​boost

​coalesce

​Comparison operators

​eq

​ne

​is_null

​is_not_null

​gt

​gte

​lt

​lte

​starts_with

​contains

​in

​match_all

​match_any

​regexp_match

​Mathematical operators

​add

​sub

​mul

​div

​abs

​min

​max

​ln

​exp

​sqrt

​square

​Collection

​topk

​limit

​sort

​count

​Query options

​LSN-based Consistency

​How it works

Query structure

Select

Select expressions

Function expressions

Vector distance

BM25 Score

Semantic similarity

Multi-vector distance

Advanced select expressions

Filtering

Metadata filtering

Keyword search

Match multiple terms

Give weight to specific terms

Combine keyword search and metadata filtering

Operators

Logical operators

and

or

not

all

any

choose

boost

coalesce

Comparison operators

eq

ne

is_null

is_not_null

gt

gte

lt

lte

starts_with

contains

in

match_all

match_any

regexp_match

Mathematical operators

add

sub

mul

div

abs

min

max

ln

exp

sqrt

square

Collection

topk

limit

sort

count

Query options

LSN-based Consistency

How it works