kdrant-core/dev.kdrant.model

Package-level declarations

Types

AliasDescription

@Serializable

data class AliasDescription(val aliasName: String, val collectionName: String)

One alias→collection mapping, as returned by listAliases / listCollectionAliases.

AliasOperation

@Serializable(with = AliasOperationSerializer::class)

sealed interface AliasOperation

A single change applied by updateAliases. All operations in one call are applied atomically, which is what makes a zero-downtime reindex possible (build a new collection, then swap the alias).

CollectionDescription

@Serializable

data class CollectionDescription(val name: String)

A collection's name, as returned by listCollections.

CollectionInfo

@Serializable

data class CollectionInfo(val status: CollectionStatus = CollectionStatus.UNKNOWN, val pointsCount: Long? = null, val indexedVectorsCount: Long? = null, val segmentsCount: Int? = null)

Summary of a collection returned by getCollection.

CollectionStatus

@Serializable(with = CollectionStatusSerializer::class)

enum CollectionStatus : Enum<CollectionStatus>

Optimization/health status of a collection. Unrecognized values (from a newer server) decode to UNKNOWN rather than failing the whole response.

Condition

@Serializable(with = ConditionSerializer::class)

sealed interface Condition

A single filter condition. Most conditions target a payload Field with a FieldMatcher; the rest cover Qdrant's special conditions (is_empty, is_null, has_id, has_vector, nested) and recursive nesting of a whole Filter via Sub.

ContextPair

data class ContextPair(val positive: VectorInput, val negative: VectorInput)

A positive/negative example pair for discovery and context search.

CreateCollectionRequest

@Serializable

data class CreateCollectionRequest(val vectors: VectorsConfig? = null, val sparseVectors: Map<String, SparseVectorParams>? = null, val hnswConfig: HnswConfig? = null, val onDiskPayload: Boolean? = null, val shardNumber: Int? = null, val replicationFactor: Int? = null, val optimizersConfig: OptimizersConfig? = null, val quantizationConfig: QuantizationConfig? = null)

Request body for PUT /collections/{name}.

DeleteSelector

sealed interface DeleteSelector

Selects which points a delete targets: an explicit id list or a filter.

Direction

enum Direction : Enum<Direction>

Sort direction for QueryInterface.OrderBy.

Distance

@Serializable

enum Distance : Enum<Distance>

Vector distance metric. Wire values match Qdrant exactly.

FacetHit

@Serializable

data class FacetHit(val value: FacetValue, val count: Long)

A distinct value of a faceted payload key together with how many points carry it. Returned by facet, ordered by descending count.

FacetValue

@Serializable(with = FacetValueSerializer::class)

sealed interface FacetValue

A facet value: Qdrant returns a payload key's distinct values as a string, integer, or boolean.

FieldMatcher

sealed interface FieldMatcher

The matcher applied to a payload field inside a Condition.Field. Each variant maps to exactly one Qdrant condition block (match, range, values_count, geo_*), mirroring the Qdrant filter model.

Filter

@Serializable

data class Filter(val must: List<Condition>? = null, val should: List<Condition>? = null, val mustNot: List<Condition>? = null, val minShould: MinShould? = null)

A Qdrant filter: four independent, combinable clauses. Each holds a list of Conditions and clauses can nest recursively (a condition can itself be a whole Filter — see Condition.Sub), so any boolean expression is expressible.

FusionAlgorithm

enum FusionAlgorithm : Enum<FusionAlgorithm>

Fusion algorithm combining the results of several Prefetch sources.

GeoPoint

@Serializable

data class GeoPoint(val lon: Double, val lat: Double)

A geographic point. Note Qdrant's field order: longitude then latitude.

HnswConfig

@Serializable

data class HnswConfig(val m: Int? = null, val efConstruct: Int? = null, val fullScanThreshold: Int? = null, val maxIndexingThreads: Int? = null, val onDisk: Boolean? = null, val payloadM: Int? = null)

HNSW index tuning. All fields are optional; omitted ones fall back to Qdrant's defaults. Can be set collection-wide or overridden per named vector.

LookupLocation

@Serializable

data class LookupLocation(val collection: String, val vector: String? = null)

Where to look up a point's vector for SearchRequest.lookupFrom: a collection and optional vector name.

MinShould

@Serializable

data class MinShould(val conditions: List<Condition>, val minCount: Int)

The min_should clause: at least minCount of conditions must match.

Modifier

@Serializable

enum Modifier : Enum<Modifier>

Query-time modification applied to sparse-vector values.

MultiVectorComparator

@Serializable

enum MultiVectorComparator : Enum<MultiVectorComparator>

How multi-vectors are compared.

MultiVectorConfig

@Serializable

data class MultiVectorConfig(val comparator: MultiVectorComparator)

Config enabling multi-vector (late-interaction / ColBERT) storage on a dense vector.

OptimizersConfig

@Serializable

data class OptimizersConfig(val defaultSegmentNumber: Int? = null, val indexingThreshold: Int? = null, val memmapThreshold: Int? = null, val flushIntervalSec: Int? = null, val maxOptimizationThreads: Int? = null, val deletedThreshold: Double? = null, val vacuumMinVectorNumber: Int? = null, val maxSegmentSize: Int? = null)

Optimizer tuning for a collection. All fields are optional; on create, omitted ones use Qdrant's defaults; on update, omitted ones keep the current value.

Payload

typealias Payload = JsonObject

A point payload: an arbitrary JSON object with heterogeneous values (strings, numbers, booleans, arrays, nested objects). Kept as a JsonObject rather than a fixed schema, since Qdrant payloads are schemaless; typed access is left to the caller.

PayloadSchemaType

@Serializable

enum PayloadSchemaType : Enum<PayloadSchemaType>

Payload field index type for createPayloadIndex. Indexing a field makes filtering on it scale.

PointGroup

@Serializable

data class PointGroup(val id: JsonPrimitive, val hits: List<ScoredPoint>, val lookup: Record? = null)

A group of hits sharing the same group_by value, returned by searchGroups.

PointId

@Serializable(with = PointIdSerializer::class)

sealed interface PointId

A Qdrant point identifier. Qdrant accepts exactly two forms: an unsigned 64-bit integer or a UUID string. Modelled as a sealed type so the two cases are impossible to confuse.

PointStruct

@Serializable

data class PointStruct(val id: PointId, val vector: VectorData, val payload: Payload? = null)

A single point to upsert: an id, its vector, and an optional payload.

PointVectors

@Serializable

data class PointVectors(val id: PointId, val vector: VectorData)

A point's id and the vector(s) to write, for updateVectors.

Prefetch

@Serializable

data class Prefetch(val prefetch: List<Prefetch>? = null, val query: QueryInterface? = null, val using: String? = null, val filter: Filter? = null, val limit: Int? = null, val scoreThreshold: Double? = null, val params: SearchParams? = null, val lookupFrom: LookupLocation? = null)

A prefetch sub-request: candidates fetched first, then combined or reranked by the outer SearchRequest.query. Prefetches nest recursively for multi-stage retrieval.

QuantizationConfig

@Serializable(with = QuantizationConfigSerializer::class)

sealed interface QuantizationConfig

Vector quantization, to shrink a collection's memory footprint. Build with Scalar or Binary.

QueryInterface

@Serializable(with = QueryInterfaceSerializer::class)

sealed interface QueryInterface

The query of a /points/query request: what to do with the (optionally prefetched) candidates.

RecommendStrategy

enum RecommendStrategy : Enum<RecommendStrategy>

How QueryInterface.Recommend uses positive/negative examples. Default (null) is average-vector.

Record

@Serializable

data class Record(val id: PointId, val payload: Payload? = null, val vector: VectorData? = null)

A point returned by scroll (no similarity score).

ScoredPoint

@Serializable

data class ScoredPoint(val id: PointId, val score: Float, val version: Long? = null, val payload: Payload? = null, val vector: VectorData? = null)

A search hit: a point with its similarity score.

ScrollPage

@Serializable

data class ScrollPage(val points: List<Record>, val nextPageOffset: PointId? = null)

One page of scroll results plus the cursor to the next page.

ScrollRequest

@Serializable

data class ScrollRequest(val filter: Filter? = null, val limit: Int, val offset: PointId? = null, val withPayload: WithPayload? = null, val withVector: Boolean? = null)

Request body for a single POST /collections/{name}/points/scroll page.

SearchGroupsRequest

@Serializable

data class SearchGroupsRequest(val groupBy: String, val groupSize: Int? = null, val limit: Int? = null, val prefetch: List<Prefetch>? = null, val query: QueryInterface? = null, val using: String? = null, val filter: Filter? = null, val params: SearchParams? = null, val scoreThreshold: Double? = null, val withPayload: WithPayload? = null, val withVector: Boolean? = null, val lookupFrom: LookupLocation? = null)

Request body for POST /collections/{name}/points/query/groups.

SearchMatrixOffsets

@Serializable

data class SearchMatrixOffsets(val offsetsRow: List<Long>, val offsetsCol: List<Long>, val scores: List<Float>, val ids: List<PointId>)

Distance matrix in sparse coordinate (COO) form: scores[i] is the score between the points at ids[offsetsRow[i]] and ids[offsetsCol[i]]. Compact for feeding clustering / dimensionality reduction.

SearchMatrixPair

@Serializable

data class SearchMatrixPair(val a: PointId, val b: PointId, val score: Float)

A single edge of the distance matrix: point a to point b with their similarity score.

SearchMatrixPairs

@Serializable

data class SearchMatrixPairs(val pairs: List<SearchMatrixPair>)

Distance matrix in pairs form: an explicit list of scored point pairs.

SearchMatrixRequest

@Serializable

data class SearchMatrixRequest(val filter: Filter? = null, val sample: Int? = null, val limit: Int? = null, val using: String? = null)

Request body for the distance-matrix analytics endpoints. Qdrant samples sample points and, for each, finds its limit nearest neighbours among the sample, optionally restricted by filter and computed on the using named vector.

SearchParams

@Serializable

data class SearchParams(val hnswEf: Int? = null, val exact: Boolean? = null, val indexedOnly: Boolean? = null)

Fine-tuning for a search's accuracy/speed trade-off.

SearchRequest

@Serializable

data class SearchRequest(val prefetch: List<Prefetch>? = null, val query: QueryInterface? = null, val using: String? = null, val filter: Filter? = null, val limit: Int? = null, val offset: Int? = null, val withPayload: WithPayload? = null, val withVector: Boolean? = null, val scoreThreshold: Double? = null, val params: SearchParams? = null, val lookupFrom: LookupLocation? = null)

Request body for POST /collections/{name}/points/query.

SnapshotDescription

@Serializable

data class SnapshotDescription(val name: String, val creationTime: String? = null, val size: Long, val checksum: String? = null)

Metadata for one snapshot, returned by the create and list operations.

SnapshotPriority

@Serializable

enum SnapshotPriority : Enum<SnapshotPriority>

Source-of-truth policy when recovering a snapshot into a collection that has replicas. Ignored for a single-replica collection.

SparseVectorParams

@Serializable

data class SparseVectorParams(val modifier: Modifier? = null)

Config for a named sparse vector under CreateCollectionRequest.sparseVectors.

UpdateCollectionRequest

@Serializable

data class UpdateCollectionRequest(val optimizersConfig: OptimizersConfig? = null, val hnswConfig: HnswConfig? = null, val quantizationConfig: QuantizationConfig? = null)

Request body for PATCH /collections/{name} — tune an existing collection's config.

VectorData

@Serializable(with = VectorDataSerializer::class)

sealed interface VectorData

The vector(s) attached to a point, in any shape Qdrant accepts or returns:

VectorDatatype

@Serializable

enum VectorDatatype : Enum<VectorDatatype>

Storage datatype for vector elements. Defaults to FLOAT32 server-side when omitted.

VectorInput

sealed interface VectorInput : QueryInterface

A bare vector input — usable directly as a nearest-search QueryInterface and as a positive/negative example inside QueryInterface.Recommend, QueryInterface.Discover and ContextPair. One of a dense QueryInterface.Vector, a stored-point QueryInterface.ById, a QueryInterface.Sparse, or a QueryInterface.MultiVector.

VectorParams

@Serializable

data class VectorParams(val size: Long, val distance: Distance, val onDisk: Boolean? = null, val datatype: VectorDatatype? = null, val hnswConfig: HnswConfig? = null, val multivectorConfig: MultiVectorConfig? = null)

Configuration for a single (dense) vector: dimensionality, metric, and storage options.

VectorsConfig

@Serializable(with = VectorsConfigSerializer::class)

sealed interface VectorsConfig

The vectors config of a collection, which Qdrant represents in two shapes:

WithPayload

@Serializable(with = WithPayloadSerializer::class)

sealed interface WithPayload

Payload projection for search/scroll results. Qdrant accepts four wire shapes: