Qdrant

1. Overview

The Qdrant node in GoInsight allows you to seamlessly integrate vector database operations into your automated workflows. Qdrant is a high-performance, scalable vector database and vector similarity search engine, essential for building AI applications, semantic search, and recommendation systems.

Through this node, you can manage the full lifecycle of your vector data and collections, including:

• Creating, updating, and deleting vector collections and their configurations.
• Managing points and vectors, including upserting, retrieving, and deleting data.
• Performing advanced vector searches such as nearest neighbor, recommendation, and batch queries.
• Handling payload (metadata) by setting, clearing, and indexing payload fields for efficient filtering.

2. Prerequisites

Before using the Qdrant node, you need to meet the following requirements:

• Qdrant Instance: You must have access to a running Qdrant instance (either Qdrant Cloud or a self-hosted instance).
• API Key: If your Qdrant instance is secured, you will need a valid API Key to authenticate your requests.

3. Credentials

For detailed instructions on how to obtain and configure credentials, please refer to our official documentation: Credentials Configuration Guide.

4. Supported Operations

Summary

This node primarily operates on the following resources: Collections, Points, Vectors, and Payloads.

Operation Details

Collection Exists

Check if a specific collection exists in Qdrant. Use this to verify collection availability before search or creation operations.

Required: CollectionName - target collection name.

Returns: Exists (bool) - true if found, false if missing; CollectionName (string) - echo of input; Time (number) - execution seconds. Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Uses HTTP HEAD request (lightweight metadata check). Returns Exists=false (not an error) if collection is missing (404).

Limitations: Qdrant_URL must be the base URL (e.g., 'http://localhost:6333') without '/collections' suffix.

Input Parameters:

• CollectionName: Target collection name to check for existence. Must be a valid Qdrant collection name (alphanumeric characters, underscores, and hyphens allowed, case-sensitive). Typical naming conventions use snake_case or kebab-case. Example: "user_embeddings_v2", "product_vectors", "documents-2024"

Output:

• Exists (bool): Whether the collection exists. This is the primary result field. True = collection found, False = collection not found (404).
• CollectionName (string): Echo of the checked collection name.
• Time (number): Operation execution time in seconds.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for exists, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Create Collection

Create a new vector collection in Qdrant. Use this to initialize a vector store; for adding points use 'Upsert Points'.

Required: CollectionName - unique identifier; VectorsConfig - size and distance metric.

Optional: AdditionalConfigs - advanced tuning (ShardNumber, ReplicationFactor, OnDiskPayload, HNSW, etc.); Timeout (default: 30).

Returns: Result (bool), Status (string), Time (number), VectorsConfig (object). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Creates a collection with specified configuration. If collection exists, returns 409.

Limitations: Collection name must be unique.

Input Parameters:

• CollectionName: Name of the collection to create. Must be unique within the Qdrant instance. Example: "my_vector_collection"
• VectorsConfig: Configuration for the collection vectors. Must include 'size' (integer) and 'distance' (string: 'Cosine', 'Euclidean', 'Dot', or 'Manhattan'). Example: {"size": 1536, "distance": "Cosine"}

Options:

• AdditionalConfigs: ⚠️ Expert-only: Advanced tuning parameters. Supported keys: ShardNumber (integer, default 1), ReplicationFactor (integer, default 1), WriteConsistencyFactor (integer, default 1), OnDiskPayload (boolean, default true), hnsw_config (object), wal_config (object), quantization_config (object). Example: {"ShardNumber": 2, "OnDiskPayload": true}
• Timeout: Request timeout in seconds for the creation process. Default is 30.

Output:

• Result (bool): Whether the operation was successful.
• Status (string): Current status of the collection (e.g., 'green', 'yellow').
• Time (number): Time taken for the operation in seconds.
• VectorsConfig (object): The configuration used for vectors. Access sub-fields via VectorsConfig.size and VectorsConfig.distance.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 409 for conflict).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description if operation fails, empty string on success.

Delete Collection

⚠️ WARNING: IRREVERSIBLE OPERATION

Delete a collection from Qdrant. This will PERMANENTLY remove the collection and ALL associated vectors/metadata. This operation CANNOT be undone.

Required: CollectionName - name of the collection to delete.

Optional: Timeout - operation timeout in seconds (default: 30).

Returns: Result (bool) - API flag; Status (string) - operation status; Deleted (bool) - confirmation flag; Time (number) - execution time; CollectionName (string). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Returns 'not_found' status if collection does not exist.

Input Parameters:

• CollectionName: The name of the collection to PERMANENTLY delete. This operation is irreversible. Collection names are case-sensitive. Example: "my_documents_v2"

Options:

• Timeout: Operation timeout in seconds. Controls how long the Qdrant server will wait before aborting the deletion. For collections with millions of vectors, recommend 60+ seconds. Default: 30. Example: 60

Output:

• Result (bool): Internal boolean flag returned by the Qdrant API indicating if the request was accepted and processed. For definitive confirmation of collection removal, refer to the 'Deleted' field. Example: true
• Status (string): The operational status of the request. Possible values: 'ok' (successfully processed), 'error' (server-side or logic error), 'not_found' (the specified collection does not exist). Always check 'ErrorMessage' when status is not 'ok'. Example: "ok"
• Time (number): Operation execution time in seconds.
• CollectionName (string): Name of the deleted collection (echoes input).
• Deleted (bool): Primary confirmation flag. Returns true only if the collection was successfully and permanently removed from the cluster. If the collection did not exist or deletion failed, this will be false. Example: true
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Get Collection

Get detailed information about a specific collection in Qdrant. Use this to retrieve configuration, status, and statistics; for listing all collections use 'List Collections'.

Required: CollectionName - the name of the collection.

Returns: CollectionInfo (object) - detailed collection data. Key fields: status (string), vectors_count (number), segments_count (number), config.params (object), config.hnsw_config (object). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Limitations: Returns 404 if the collection does not exist.

Input Parameters:

• CollectionName: Name of the collection to retrieve information for. A collection in Qdrant is a container for storing vector data, similar to a table in a traditional database. Collection names are case-sensitive and must be unique within the Qdrant instance. Example: "user_embeddings"

Output:

• CollectionInfo (object): Detailed collection information object containing: - status (string): Collection health status (green/yellow/red) - optimizer_status (string): Optimizer state (ok/optimizing) - vectors_count (number): Total number of vectors in collection - indexed_vectors_count (number): Number of indexed vectors - points_count (number): Total number of points - segments_count (number): Number of segments - config (object): Collection configuration including: - params: Vector parameters (size, distance metric) - hnsw_config: Hierarchical Navigable Small World index settings for efficient vector search - optimizer_config: Settings for background data optimization - wal_config: Write-Ahead Log settings for data durability - payload_schema (object): Schema definitions for metadata (payload) fields Example: {"status": "green", "vectors_count": 15420, "config": {...}}
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

List Collections

List all collections in a Qdrant vector database. Use this to retrieve collection names and configurations.

⚠️ WARNING: Qdrant API does not support pagination; returns ALL collections in one request. Suitable for small clusters (<100 collections).

Returns: Collections (object) - wrapper containing list and metadata. Key fields: collections (array), total_count (int), has_more (bool). Plus Result, Status, Time, StatusCode, ErrorMessage, OriginalStatusCode.

Limitations: 401 if API Key is invalid; 500 if server unreachable. Large clusters may consume significant tokens.

Output:

• Result (bool): Boolean indicating whether the operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• Collections (object): The response containing the list of collections and metadata. Key fields: - collections (array): List of collection objects (each contains 'name'). - total_count (number): Total number of collections in the cluster. - has_more (boolean): Always false as Qdrant returns all collections in one request. Example access: Collections.collections[0].name, Collections.total_count
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 401 for unauthorized). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Update Collection

Update the configuration of an existing Qdrant collection. Use this to modify settings like optimizers or HNSW index parameters; for creating new collections use 'Create Collection'.

Required: CollectionName - the name of the collection to update.

Optional: OptimizersConfig - indexing and cleanup settings; HnswConfig - index accuracy vs speed settings; QuantizationConfig - vector compression settings; AdditionalConfig - advanced settings (params, sparse_vectors, strict_mode_config, vectors); Timeout - operation timeout in seconds (default: 30).

Returns: UpdatedConfig (object) - the modified configuration. Key fields: optimizers_config (object), hnsw_config (object), params (object). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: This is an incremental update; only the specified fields will be modified. At least one configuration parameter must be provided.

Limitations: Returns 404 if the collection does not exist.

Input Parameters:

• CollectionName: The unique name of the collection to update. Example: "my_vector_store"

Options:

• OptimizersConfig: Optimizer configuration. Controls background processes that manage indexing and memory optimization. Supported keys: indexing_threshold (number) - vector count before indexing starts (default 20000); memmap_threshold (number) - threshold for memory mapping; deleted_threshold (number) - ratio of deleted vectors to trigger cleanup. Example: {"indexing_threshold": 10000, "memmap_threshold": 50000}
• HnswConfig: HNSW (Hierarchical Navigable Small World) index configuration. Controls search accuracy vs speed tradeoff. Supported keys: m (number) - edges per node (4-64); ef_construct (number) - candidate list size during construction; full_scan_threshold (number) - switch to brute-force search below this count. Example: {"m": 16, "ef_construct": 100}
• QuantizationConfig: Vector compression settings (Quantization). Reduces memory usage by up to 4x with minimal accuracy loss. Supported keys: scalar (object) - scalar quantization settings (e.g., {"type": "int8", "quantile": 0.99}); product (object) - product quantization settings. Example: {"scalar": {"type": "int8", "quantile": 0.99}}
• AdditionalConfig: Advanced configuration grouping. Use for low-frequency updates. Supported keys: params (object) - basic parameters like replication_factor; sparse_vectors (object) - sparse vector settings; strict_mode_config (object) - strict mode constraints; vectors (object) - general vector configuration updates. Example: {"params": {"replication_factor": 2}}
• Timeout: Operation timeout in seconds. Increase for large collections. Example: 60

Output:

• Result (bool): Indicates if the update operation was successful.
• Status (string): Status message from the server (e.g., 'ok').
• Time (number): Time taken for the operation in seconds.
• UpdatedConfig (object): Object containing the modified configuration fields. Key fields: optimizers_config (object), hnsw_config (object), quantization_config (object), params (object).
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Matrix Offsets

Compute similarity offset matrix between sample points and collection points in Qdrant. Returns results in compressed sparse format (CSR) compatible with scipy.

Required: CollectionName (from 'List Collections'); Sample (number of points to randomly sample, must be >= 2).

Optional: Limit (items per result, default: 10); WithPayload/WithVectors/Consistency.

Returns: MatrixOffsets (object with offsets_row, offsets_col, scores, ids), SampleConfig, StatusCode, ErrorMessage, OriginalStatusCode.

Note: This endpoint returns a single similarity matrix in CSR format, not multiple offset-based paginations. For pagination, use the standard search endpoint with offset parameters.

Input Parameters:

• CollectionName: Name of the collection to search in. Can be obtained via List Collections action. Example: "my_products"
• Sample: Number of points to randomly sample from the collection. Must be >= 2. This determines how many reference points will be used to compute the similarity matrix. Example: 10

Options:

• Limit: Max number of neighbors to find for each sampled point. Example: 10
• WithPayload: Whether to return payload data (business metadata like product name, price, category) with each point.
• WithVectors: Whether to return vector embeddings. Vectors are the numerical representations (DNA) of items. Turn ON to re-sort results (re-rank) based on full vector data or for debugging; turn OFF (default) to return business metadata only. Example: false
• Consistency: Read consistency level (Data freshness guarantee). Controls speed vs accuracy:\n- majority (recommended): Balanced, waits for most nodes to agree.\n- quorum: Most accurate, waits for strict consensus.\n- all: All nodes must agree.\n- Empty string (default): Uses Qdrant's default.\nLeave empty for best performance.

Output:

• MatrixOffsets (object): Similarity matrix result in compressed sparse format (CSR). Contains:\n- offsets_row (array): Row indices of non-zero similarities.\n- offsets_col (array): Column indices of non-zero similarities.\n- scores (array): Similarity scores (0.0-1.0) at corresponding positions.\n- ids (array): Point IDs in row/column order.\nThis format is compatible with scipy's csr_matrix for further analysis.
• SampleConfig (object): The sampling configuration used for this request. Contains:\n- sample (int): Number of points randomly sampled.\n- limit (int): Max results per query.\nExample: {\"sample\": 10, \"limit\": 10}
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Matrix Pairs

Compare two groups of points (Group A and Group B) within a Qdrant collection to find similarities. Use this for recommendation systems, cross-category analysis, and finding relationships between different data subsets.

Required: CollectionName - the name of the Qdrant collection; Sample - number of points to randomly sample (integer, must be >= 2).

Optional: Filters - filter conditions for Group A and Group B; Limit - max points per group (default: 10); WithPayload - include metadata; WithVectors - include vectors; Consistency - read consistency level (majority/quorum/all, leave empty for default).

Returns: MatrixPairs (object-array) - list of comparison results; SampleConfig (object) - effective sampling config. Plus Result, Status, Time, PairsCount, GroupASize, GroupBSize, StatusCode, ErrorMessage, OriginalStatusCode.

Terminology:

• Distance Matrix: A table showing the similarity score between every point in Group A and every point in Group B.
• Payload: Metadata attached to points.
• Vector: The mathematical representation of data.

Input Parameters:

• CollectionName: The name of the Qdrant collection to perform the matrix analysis on. Example: "my_vectors"
• Sample: Number of points to randomly sample from the collection. Must be >= 2. This determines how many reference points will be used to compute similarity pairs. Example: 10

Options:

• Filters: Filter conditions for the two groups of points to be compared. Contains two optional keys: - a (object): Filter for Group A. - b (object): Filter for Group B. Each filter uses Qdrant syntax: - must: All conditions must match (AND). - should: At least one condition matches (OR). - must_not: Conditions must not match (NOT). Example: { "a": { "must": [{"key": "category", "match": {"value": "electronics"}}] }, "b": { "must": [{"key": "category", "match": {"value": "books"}}] } }
• Limit: Number of points to return per group. Default is 10.
• WithPayload: Whether to include 'Payload' (the non-vector metadata/attributes associated with each point, like name or category) in the result. Set to true if you need to identify points by their attributes.
• WithVectors: Whether to include the raw 'Vectors' (the numerical embedding arrays) in the result. Usually set to false unless you need the raw data for further local computation.
• Consistency: Read consistency level for the operation. Choose based on your performance vs. accuracy needs:\n- majority: Returns data confirmed by most nodes (recommended, balanced).\n- quorum: Highest accuracy, ensures data is consistent across a quorum of nodes.\n- all: All nodes must agree (slowest, most accurate).\n- Empty string (default): Uses Qdrant's default consistency.\nLeave empty for best performance.

Output:

• Result (bool): Boolean indicating whether the matrix pairs computation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where matrix pairs were computed.
• MatrixPairs (object-array): A list of comparison results between points from Group A and Group B. Each object contains: - a (object): Details of the point from Group A (id, score, and optional payload). - b (object): Details of the point from Group B (id, score, and optional payload). - score (float): The similarity score between point A and point B (typically 0 to 1).
• PairsCount (number): Total number of point pairs returned in the matrix computation.
• GroupASize (number): Number of unique points in the first group (group A).
• GroupBSize (number): Number of unique points in the second group (group B).
• SampleConfig (object): The effective sampling configuration applied to the request. Contains:\n- sample (int): Number of points randomly sampled.\n- limit (int): Max results per query.\nExample: {\"sample\": 10, \"limit\": 10}
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description if operation fails, empty string on success.

Clear Payload

Clear all payload (metadata) from specified points in a Qdrant collection. Use this to remove metadata while preserving vector embeddings.

Required: CollectionName - the collection name.

Optional: Points - list of point IDs (UUIDs or integers); FilterCondition - Qdrant filter object (mutually exclusive with Points); Wait - wait for completion (default: true); Ordering - consistency level (values: weak/medium/strong, default: weak); ShardKey - custom shard key.

Returns: Result (bool), Status (string), Time (number), ClearedPointsCount (number), OperationId (number), OperationStatus (string), Usage (object). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Irreversible operation. Either Points or FilterCondition must be provided, but not both. If Wait is true, returns after operation commit.

Limitations: 404 if collection not found; 400 if parameters are invalid.

Input Parameters:

• CollectionName: Name of the Qdrant collection to modify. Example: "my_documents"

Options:

• Points: List of specific point IDs to clear payload from. Each ID can be a UUID string or an integer. Mutually exclusive with FilterCondition. Example: ["550e8400-e29b-41d4-a716-446655440000", 123]
• FilterCondition: Filter condition to clear payload from points matching the criteria. Mutually exclusive with Points. Use Qdrant filter syntax. Example: {"must": [{"key": "status", "match": {"value": "archived"}}]}
• Wait: If true (default), the request will wait until the operation is committed to all replicas. If false, it returns immediately with an operation ID.
• Ordering: Write ordering consistency level. Affects operation completion time and data reliability. Values: 'weak' (default): Fastest, suitable for most scenarios where eventual consistency is acceptable; 'medium': Balanced consistency; 'strong': Most reliable but slowest, use only for strict consistency requirements (e.g., financial data). Recommendation: Keep default 'weak' unless strong consistency is required. Example: "weak"
• ShardKey: Data sharding strategy key. ⚠️ Keep empty for most users. Only use if you have a custom sharding setup and need to target a specific shard. Example: "shard_1"

Output:

• Result (bool): Boolean indicating whether the payload clearing operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where payload clearing operation was performed.
• ClearedPointsCount (number): Number of points that had their payload data cleared successfully. When using Points parameter, this equals the number of IDs provided. When using FilterCondition, this may be 0 if the Qdrant API version does not return the affected count. Example: 5
• OperationId (number): Operation ID for asynchronous execution tracking.
• OperationStatus (string): Current status of the operation if available.
• Usage (object): Resource usage statistics for the operation. This is a JSON object with the following structure: {"cpu": 1250, "payload_io_read": 10, "payload_io_write": 5, "payload_index_io_read": 0, "payload_index_io_write": 0, "vector_io_read": 0, "vector_io_write": 0}. All values are integers representing micro-units or operation counts.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Delete Payload

Delete specific payload keys from selected points in a Qdrant vector database collection. Use this to remove metadata fields while keeping points; for deleting points use 'Delete Points'.

Required: CollectionName - target collection; Keys - list of field names to delete.

Optional: Points - list of point IDs (mutually exclusive with FilterCondition); FilterCondition - Qdrant filter object; Wait - wait for disk commit (default: true); Ordering - consistency level (weak/medium/strong); ShardKey - for custom sharding.

Returns: Result (bool), Status (string), Time (number), Usage (object) - resource consumption stats. Key fields: cpu (int), payload_io_read (int). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Returns immediately if Wait is false.

Limitations: Cannot provide both Points and FilterCondition.

Input Parameters:

• CollectionName: Name of the collection to delete payload from. Example: "my_collection"
• Keys: List of specific payload field names to delete from the points. Example: ["metadata", "tags"]

Options:

• Points: WARNING: Mutually exclusive with FilterCondition. You MUST provide either Points or FilterCondition (choose one). List of specific point IDs to delete payload from. Can be integers or strings (UUIDs). Use this when you know the exact point IDs. Example: [1, 2, "uuid-abc-123"] Note: If both Points and FilterCondition are provided, the request will fail with error -1.
• FilterCondition: WARNING: Mutually exclusive with Points. You MUST provide either Points or FilterCondition (choose one). Qdrant filter condition to select points for payload deletion. Use this to target points matching specific criteria. Filter structure: - must: All conditions must match (AND logic) - should: At least one condition must match (OR logic) - must_not: Conditions must not match (NOT logic) Example - Delete from points in London: { "must": [ { "key": "city", "match": { "value": "London" } } ] } Example - Multiple conditions: { "must": [ { "key": "status", "match": { "value": "active" } }, { "key": "score", "range": { "gte": 0.8 } } ] } Note: If both Points and FilterCondition are provided, the request will fail with error -1.
• Wait: Whether to wait for the operation to complete. Synchronous if true, asynchronous if false. Example: true
• Ordering: Write consistency level. Controls the trade-off between speed and data consistency: - weak (default): Fastest. Changes may not be immediately visible to all nodes. - medium: Balanced. Changes visible after acknowledgment from majority of nodes. - strong: Slowest. Changes immediately visible to all nodes. Recommendation: Use 'weak' for most scenarios. Use 'strong' only if you need immediate consistency across all nodes. Example: "weak"
• ShardKey: Advanced Option: Shard key for data distribution control. In a multi-node Qdrant cluster, this determines which server stores your data. Only use this if you have a custom sharding strategy configured. For single-node deployments, leave this empty. Example: "user_group_1"

Output:

• Result (bool): Boolean indicating whether the payload key deletion operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where payload keys were deleted.
• DeletedKeys (string-array): Array of payload key names that were deleted from the selected points.
• OperationId (number): Operation ID for tracking asynchronous operations (0 if synchronous).
• OperationStatus (string): Status of the payload key deletion operation if available.
• Usage (object): Resource usage statistics for the operation. Contains the following integer fields: - cpu: CPU time consumed (microseconds) - payload_io_read: Number of payload read operations - payload_io_write: Number of payload write operations - payload_index_io_read: Number of payload index read operations - payload_index_io_write: Number of payload index write operations - vector_io_read: Number of vector read operations - vector_io_write: Number of vector write operations Example: {"cpu": 1250, "payload_io_read": 5, "payload_io_write": 2, "payload_index_io_read": 0, "payload_index_io_write": 0, "vector_io_read": 0, "vector_io_write": 0}
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Overwrite Payload

Overwrite payload data for specific points in a Qdrant collection. Use this to completely replace existing payload (PUT behavior); for partial updates (PATCH), use 'Set Payload'.

Required: CollectionName - collection name; Payload - new data.

Optional: Points - target IDs; FilterCondition - target filter; Overwrite - replace mode (default: true); Key - specific path; Wait - wait for commit; Ordering - consistency level (values: weak/medium/strong, default: weak); ShardKey - distributed routing.

Returns: Result (bool), Status (string), PayloadData (object), Usage (object). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: None.

Behavior: Destructive operation: fields not in new Payload are deleted if Overwrite is true. Use Cases: Overwrite resets all metadata; Set Payload updates specific fields.

Limitations: Points and FilterCondition are mutually exclusive.

Input Parameters:

• CollectionName: Name of the collection where points are stored. Example: "customer_profiles"
• Payload: The payload data to write. If 'Key' is provided, this object is written to that specific path. If 'Key' is omitted, this object replaces the entire payload. Example: {"status": "active", "score": 100}

Options:

• Points: List of point IDs (strings or integers) to overwrite payload for. Example: [1, 2, "uuid-3"]. Note: This parameter is mutually exclusive with FilterCondition; if both are provided, Points takes precedence.
• FilterCondition: Filter criteria to select points for payload overwrite. Use this to target points based on their existing metadata. Example: {"must": [{"key": "city", "match": {"value": "London"}}]}. Note: This parameter is mutually exclusive with Points.
• Overwrite: Control overwrite behavior. If true (default), completely replace existing payload at the target location (deleting old fields). If false, merge new payload with existing data. Default: true.
• Key: Optional path to a specific field in the payload to overwrite (e.g., "metadata.tags"). If not provided, the entire payload object will be replaced by the new payload. Leave empty to overwrite the root object.
• Wait: If true, the API waits for the operation to be committed to disk before returning. Recommended for consistency.
• Ordering: Data consistency level, affecting the balance between speed and reliability. Values: 'weak' (default) - Fastest. Data is written immediately but might be lost in rare server crashes. Best for logs or temporary data. 'medium' - Balanced. Waits for the main server to confirm the write. Suitable for most business data. 'strong' - Slowest but most reliable. Waits for all backup servers (replicas) to confirm. Use for critical data where loss is unacceptable.
• ShardKey: Advanced: Distributed routing key. Only required for large-scale deployments that use custom data partitioning (sharding). What is a shard? In distributed databases, data is split into multiple partitions (shards) for parallel processing. This parameter specifies which partition to route the request to. If you are unsure whether you need this, leave it empty. Single-node and standard cloud users typically do not need this.

Output:

• Result (bool): Boolean indicating whether the payload overwrite operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where payload was overwritten.
• PayloadData (object): Empty object. To retrieve the actual stored payload after this destructive operation, use the 'Retrieve Points' action.
• OperationId (number): Operation ID for tracking the payload overwrite process.
• OperationStatus (string): Current status of the payload overwrite operation.
• Usage (object): Detailed resource usage statistics for the operation. All returned fields are of type number. Common fields include: cpu_time (milliseconds), payload_index_io_read, payload_index_io_write, payload_io_read, payload_io_write, vector_io_read, vector_io_write. Note: The specific fields present may vary based on the operation complexity and server version.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description. Empty string if successful.

Payload Facets

Retrieve facet statistics for a payload field in a Qdrant collection. Use this to analyze data distribution (e.g., count items by category); for searching points use 'Search Points'.

Required: CollectionName - target collection; Key - payload field name (e.g., 'color', 'size').

Optional: Limit (default: 10); FilterCondition - JSON object to filter points; Exact (default: false) - true for precise counts; ShardKey - partition identifier; Consistency (values: eventual/majority/quorum, default: majority); Timeout (default: 0).

Returns: Facets (object-array) - list of values and counts. Key fields: value (string/number), count (number). Plus Usage, Result, Time, StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Returns unique values and their occurrence counts. Supports approximate counting for performance.

Limitations: High cardinality fields may be slow with exact=true.

Input Parameters:

• CollectionName: Target collection name (required). Example: "product_catalog"
• Key: Payload field name for facet analysis (e.g., category, status, type). Example: "category"

Options:

• Limit: Maximum number of facet values to return for the key. Must be a positive integer. Default: 10. Example: 20
• FilterCondition: Filter criteria to narrow down the points before performing facet analysis. Supports Qdrant filter syntax. Example: {"must": [{"key": "active", "match": {"value": true}}]}
• Exact: Whether to perform exact counting (true for exact, false for approximate). Default: false.
• ShardKey: The identifier used to group data into logical partitions (shards). Use this to restrict the search to a specific subset of data, like a specific 'tenant_id' or 'user_group'. Example: "tenant_123"
• Timeout: Maximum time in seconds to wait for the request to complete. Default: 0 (use system default). Example: 30
• Consistency: Defines the data consistency level (how many data copies must agree). Options: 'eventual' (Fastest, may return slightly outdated data), 'majority' (Balanced, waits for more than half of data copies to agree), 'quorum' (Strongest, waits for a specific number of copies to ensure highest accuracy). Default: 'majority'. Example: "majority"

Output:

• Result (bool): Boolean indicating whether the facet analysis operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where facet analysis was performed.
• FieldKey (string): Payload field name that was analyzed for facet statistics.
• Facets (object-array): List of facet results for the specified key. Each object contains: value (string/number): The unique value found in the payload. The data type (string or number) will match the original type stored in the collection's payload; count (number): The number of points containing this specific value.
• TotalFacets (number): Total number of facets returned in the analysis results.
• IsExactCount (bool): Boolean indicating whether the facet counts are exact or approximate.
• FilterApplied (bool): Boolean indicating whether a filter condition was applied during facet analysis.
• Usage (object): Detailed resource usage for the request. Includes: cpu (number): CPU time used in microseconds (µs); payload_io_read (number): Number of payload read operations; payload_index_io_read (number): Number of index read operations.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Set Payload

Set payload (metadata) for specific points in a Qdrant collection. Use this to update metadata for existing points; for creating new points use 'Upsert Points'.

Required: CollectionName - collection name; Payload - metadata key-value pairs.

Optional: Points - list of IDs (mutually exclusive with FilterCondition); FilterCondition - filter object; AdditionalOptions - advanced settings (wait, ordering, shard_key, key).

Returns: Result (object) - complete Qdrant API response. Key fields: result.status (string), result.operation_id (int), time (number). Usage (object) - resource stats. Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Only modifies existing points. If 'key' is provided in AdditionalOptions, updates that specific path; otherwise merges the payload. If a provided Point ID does not exist, it is silently ignored.

Limitations: Cannot provide both 'Points' and 'FilterCondition'.

Input Parameters:

• CollectionName: Name of the collection where the payload will be updated. Example: "my_collection"
• Payload: The key-value pairs to be added or updated in the points' metadata. Example: {"city": "London", "tags": ["new"]}

Options:

• Points: List of specific point IDs (string or integer) to update. Mutually exclusive with FilterCondition. Example: [1, 2, "uuid-123"]
• FilterCondition: Qdrant filter object used to select a batch of points based on criteria. Structure: 'must' (AND), 'should' (OR), 'must_not' (NOT). Each contains conditions like 'match' (exact value) or 'range' (numeric limits). Mutually exclusive with Points. Example: {"must": [{"key": "city", "match": {"value": "London"}}]}
• AdditionalOptions: Advanced execution parameters. Supported keys: 'wait' (boolean, default true): Whether to wait for the operation to complete; 'ordering' (string): Consistency level ('weak': fastest, 'medium': balanced, 'strong': most reliable); 'shard_key' (string/integer): Shard key for distributed deployments to route the request; 'key' (string): Specific path in payload to update (if provided, only this path is updated instead of merging the whole object). Example: {"wait": true, "ordering": "strong"}

Output:

• Result (object): The complete Qdrant API response object. Core fields: result.operation_id (integer): The unique ID of the operation; result.status (string): Operation status, typically 'acknowledged' or 'completed'; status (string): Overall API status, usually 'ok'; time (number): Processing time in seconds. Example: {"result": {"operation_id": 10, "status": "acknowledged"}, "status": "ok", "time": 0.01}
• Usage (object): Usage statistics. Fields: status (string): Request status, e.g., 'ok'; time (number): Total execution time in seconds. Example: {"status": "ok", "time": 0.01}
• OriginalStatusCode (number): Original HTTP status code returned by the Qdrant API.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Create Payload Index

Create a payload index for a specific field in a Qdrant vector database collection. Use this to improve query performance for filtering and sorting; for creating collections use 'Create Collection'.

Required: CollectionName - target collection name; FieldName - field to index.

Optional: FieldSchema - index type (default: auto-inferred); Wait (default: true); Ordering (values: weak/medium/strong, default: weak).

Returns: Result (object) - Qdrant API response. Key fields: result (bool), status (string), time (number), usage (object, cpu/io stats). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: CollectionName from 'List Collections' action.

Behavior: Blocks until completion if Wait is true (synchronous).

Input Parameters:

• CollectionName: Name of the collection to create index for. Can be retrieved from 'List Collections' action. Example: "my_collection"
• FieldName: Name of the payload field to index. Example: "city"

Options:

• FieldSchema: Definition of the index type (JSON object). Common types: - 'keyword': For exact matches. Example: {"type": "keyword"} - 'integer' or 'float': For range queries. Example: {"type": "integer"} or {"type": "float"} - 'text': For full-text search with optional tokenizer. Example: {"type": "text", "tokenizer": "word"} If omitted, Qdrant will auto-infer the type from field values.
• Wait: Whether to wait for the operation to complete (synchronous) or return immediately (asynchronous). Default: true. Example: true
• Ordering: How quickly index changes are visible across all database nodes: - 'weak' (Recommended): Fastest. Changes appear within seconds. Good for most use cases where absolute immediate consistency is not required. - 'medium': Balanced speed and consistency. Changes appear within a few seconds. - 'strong': Slowest. Guarantees all nodes see changes immediately. Use only when critical for the next operation to see the index update. Default: "weak". Example: "weak"

Output:

• Result (object): The result of the index creation operation. Root-level fields: - 'result' (bool): Operation success flag. true = index created successfully. - 'status' (string): Operation status (e.g., 'ok'). - 'time' (float): Execution time in seconds. - 'usage' (object): Resource metrics including CPU and memory usage.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Delete Payload Index

Delete a payload index from a Qdrant collection. Use this to remove an index for a specific field; for creating an index use 'Create Payload Index'.

Required: CollectionName - Name of the collection; FieldName - Name of the payload field.

Optional: Wait (default: true); Ordering (values: weak/medium/strong, default: weak).

Returns: Result (object) - Qdrant API response. Key fields: result (bool/object), status (string), time (number). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: CollectionName from 'List Collections'.

Behavior: Permanently deletes the index. The operation removes the index to save storage space and memory usage, but may reduce query performance for the specified field.

Limitations: Irreversible.

Input Parameters:

• CollectionName: Name of the collection to delete the index from. You can use 'List Collections' to find available names. Example: "product_catalog_v2"
• FieldName: Name of the payload field to delete the index for. Example: "product_category"

Options:

• Wait: Whether to wait for the operation to complete. If true, the request will block until the index is deleted. Default is true. (Use checkbox in UI or pass boolean value)
• Ordering: Write consistency level. Controls how strictly the operation waits for data synchronization: "weak" (default, fastest), "medium" (balanced), "strong" (slowest, ensures immediate visibility). Example: "weak"

Output:

• Result (object): The complete response object returned by the Qdrant API. Key fields: result (bool/object), status (string), time (number). Refer to Qdrant API documentation for full field list.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): Operation status code: 200: Success. The upstream API request completed successfully. Check ErrorMessage for business-level errors. -1: Parameter validation error. One or more input parameters are invalid or missing. 500: System error. Network timeout, connection failure, or response parsing error. The operation may be retried.
• ErrorMessage (string): Error description if operation fails, empty string on success.

Batch Update Points

Perform batch operations on points in a Qdrant collection. Use this to execute multiple operations (upsert, delete, set_payload) in a single request; for single point operations use specific actions like 'Upsert Point'.

Required: CollectionName - target collection; Operations - list of operation objects.

Optional: Wait (default: true) - wait for completion; Ordering (values: weak/medium/strong) - consistency level.

Returns: Results (object-array) - execution details. Key fields: status (string), operation_id (int). Plus Result, Time, StatusCode, ErrorMessage, OriginalStatusCode.

Limitations: 404 if collection not found; 400 if operations format is invalid.

Input Parameters:

• CollectionName: Target collection name for batch update operations. Example: "knowledge_base"
• Operations: List of operations to perform. Supported types: 1. upsert - Insert or update points Structure: {"upsert": {"points": [{"id": , "vector": [], "payload": {: }}]}} - id: Unique point ID (string or number) - vector: Vector array (list of floats) - payload: Metadata object (optional) 2. delete - Delete points Structure: {"delete": {"points": [, , ...]}} - points: List of point IDs to delete 3. set_payload - Set metadata Structure: {"set_payload": {"payload": {: }, "points": [, ...]}} - payload: Metadata object to set - points: Target point ID list Full API Docs: https://qdrant.tech/documentation/concepts/points/#batch-update

Options:

• Wait: Whether to wait for the operation to be fully written to disk before returning. - true (default): Wait for completion, ensuring data persistence. Suitable for critical data. - false: Return immediately, operation runs asynchronously. Suitable for high throughput. Recommendation: Use default 'true' for most scenarios.
• Ordering: Data consistency level, determining the balance between speed and reliability: - weak (default): Fastest, for high-frequency writes (e.g., logs). May return success before full sync. - medium: Balanced. Ensures write to leader node. Suitable for most cases. - strong: Most reliable. Ensures write to all replicas. Suitable for critical data. Recommendation: Use 'weak' if unsure. Use 'strong' for critical data.

Output:

• Result (bool): Boolean indicating whether the batch update operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where batch operations were performed.
• OperationsCount (number): Number of operations executed in the batch request.
• Results (object-array): Array of batch operation results. Key fields: - operation_id (number): Unique operation identifier. - status (string): Operation status, e.g., 'completed', 'acknowledged' (async), 'failed'. Example: [{"operation_id": 12501, "status": "completed"}]
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Count Points

Count points in a Qdrant vector database collection. Use this to get the number of points matching specific criteria; for retrieving actual points use 'Scroll Points'.

Required: CollectionName - target collection name.

Optional: FilterCondition (object) - filter criteria (must/should/must_not logic); Exact (default: true) - true for precise count, false for faster approximate count.

Returns: Count (number) - the number of points; Result (bool) - success status; Time (number) - execution time; Status (string) - operation status. Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Approximate count (Exact=false) is significantly faster for large collections (>1M points) with small error margin.

Limitations: 404 if collection not found.

Input Parameters:

• CollectionName: Target collection name for counting points.

Options:

• FilterCondition: Filter condition object to select specific points for counting. Leave empty to count all points. Filter Structure: - must: Array of conditions that MUST match (AND logic). - should: Array of conditions where at least one MUST match (OR logic). - must_not: Array of conditions that MUST NOT match (NOT logic). Common Operators: - match: Exact match, e.g., {"key": "category", "match": {"value": "electronics"}} - range: Numeric range, e.g., {"key": "price", "range": {"gte": 100}} Example: {"must": [{"key": "status", "match": {"value": "active"}}]}
• Exact: Counting mode selection. Values: - true (Exact Count): Returns the precise number of points. Recommended for small/medium collections or critical metrics. Slower. - false (Approximate Count): Returns an estimated count. Faster (~10x) for large collections (>1M points). Default: 'true'.

Output:

• Result (bool): Boolean indicating whether the point counting operation was successful.
• Status (string): The status of the count operation. Possible values: 'ok' (success), 'error' (failure). If 'error', check ErrorMessage for details.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where point counting was performed.
• Count (number): Number of points matching the specified filter conditions.
• IsExactCount (bool): Boolean indicating whether the count is exact (true) or approximate (false).
• FilterApplied (bool): Boolean indicating whether filter conditions were applied during counting.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Delete Points

Delete points from a Qdrant vector database collection. Use this to permanently remove vectors by ID or filter criteria. ⚠️ WARNING: This operation is irreversible.

Required: CollectionName - target collection name.

Optional: Points - list of IDs to delete; FilterCondition - Qdrant filter object (must/should/must_not); Wait - execution mode (default: true); Ordering - consistency level (values: weak/medium/strong, default: weak).

Returns: Result (bool), Status (string), OperationId (number), DeletedCount (number). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Must provide either Points or FilterCondition. If Wait is false, returns OperationId for async tracking.

Limitations: Deletion cannot be undone.

Input Parameters:

• CollectionName: Target collection name.

Options:

• Points: List of point IDs to delete from the collection. Supports numeric IDs (e.g., ["101", "102"]) or UUID strings (e.g., ["550e8400-e29b-41d4-a716-446655440000"]). Note: The ID format must match the existing IDs in your collection. Example: ["1024", "1025"]
• FilterCondition: Filter condition for batch deletion. Points matching this filter will be removed. Syntax Guide: - must: All conditions must match (AND). - should: At least one condition must match (OR). - must_not: Conditions must not match (NOT). Common Operators: - match: Exact value match (e.g., {"key": "color", "match": {"value": "red"}}). - range: Numeric range (gte, lte, gt, lt). Example: {"must": [{"key": "status", "match": {"value": "archived"}}, {"key": "score", "range": {"lt": 0.5}}]} Full documentation: https://qdrant.tech/documentation/concepts/filtering/
• Wait: Execution mode. - true (Immediate): Wait for the deletion to complete across the cluster before returning. - false (Background): Return immediately while the deletion happens in the background. Example: true
• Ordering: Consistency level for the deletion. - weak: Fastest, but other nodes might see the point for a short time. - medium: Balanced performance. - strong: Ensures all nodes are updated before confirming. Use this if you need to immediately verify the deletion in a high-concurrency environment. Example: "strong"

Output:

• Result (bool): Boolean indicating whether the points deletion was successful.
• Status (string): The status of the deletion request. Possible values: 'ok' (processed), 'acknowledged' (background), 'error' (failed).
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where points were deleted.
• DeletedCount (number): Number of points successfully deleted (if available from API).
• OperationId (number): Operation ID for tracking the deletion process (if async).
• DeleteType (string): Indicates how points were identified for deletion: 'ids', 'filter', or 'mixed'.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200, 404).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Find Batch Points

Execute batch vector queries in a Qdrant collection. Use this to perform multiple searches in a single request; for single searches use 'Search Points'.

Required: CollectionName - target collection; Searches - list of query configurations.

Optional: Consistency (values: eventual/majority/quorum, default: eventual).

Returns: Results (object-array) - list of search results with metadata. Key fields: query_index (number), results (array), results_count (number). Plus Result, Status, Time, TotalResults, StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Executes all queries in a single HTTP request.

Limitations: 404 if collection not found.

Input Parameters:

• CollectionName: Target collection name
• Searches: List of search query configurations to be executed in batch. Each object in the array supports: - query (Required): Vector as [0.1, 0.2, ...] or Point ID as "uuid" or query object {"nearest": [...]}. - filter (Optional): Filter conditions for the search. - limit (Optional): Max results per query (default: 10). - with_payload (Optional): Whether to return point metadata (default: true). - score_threshold (Optional): Minimum similarity score (0.0 to 1.0). Template: [ { "query": [0.1, 0.2, 0.3], "filter": {"must": [{"key": "city", "match": {"value": "London"}}]}, "limit": 5 } ]

Options:

• Consistency: Read consistency level. Defines how many nodes must agree before returning a result. - eventual: Fastest. May return slightly stale data. Best for high-throughput recommendations. - majority: Balanced. Requires agreement from more than half of the nodes. Recommended for most production use cases. - quorum: Strongest consistency. Slowest performance. Use for critical data where accuracy is paramount. - (empty): Uses Qdrant default (usually eventual).

Output:

• Result (bool): Boolean indicating whether the batch query operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where batch queries were executed.
• SearchesCount (number): Number of search queries executed in the batch operation.
• Results (object-array): Array of results, where each element corresponds to a query in the input 'Searches' list. Structure of each result object: - query_index (number): The 0-based index of the query in the input list. - results (array): List of points found. Each point contains id, score, payload, and vector (if requested). - results_count (number): Number of points returned for this specific query.
• TotalResults (number): Total number of results returned across all queries in the batch.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Find Points

Find points in a Qdrant collection using vector similarity or other query types. Use this to search for nearest neighbors, recommendations, or discovery; for retrieving specific points by ID use 'Get Points'.

Required: CollectionName - target collection; Query - search configuration.

Optional: FilterCondition - filtering logic; Limit (default: 10); Offset (default: 0); WithPayload (default: true); WithVectors (default: false); Options - advanced settings (HNSW params, score_threshold, consistency).

Returns: Results (object-array) - list of points. Key fields: id (str/int), score (number), payload (object), vector (array). Plus Result, Status, Time, ResultsCount, SearchParams, StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: CollectionName from 'List Collections'.

Behavior: Supports complex query DSL (nearest, recommend, discover). Returns empty list if no matches found.

Limitations: 404 if collection not found.

Input Parameters:

• CollectionName: Target collection name in Qdrant. Obtain this value by calling 'List Collections' action first to see available collections. Extract the 'name' field from the returned collections array to use here. Example: "my_documents"
• Query: Search query configuration. Qdrant supports multiple query types: 1. nearest: Find similar vectors by providing a reference vector. Example: {"nearest": {"vector": [0.12, 0.34, 0.56, 0.78]}}; 2. recommend: Get recommendations based on positive/negative examples. Example: {"recommend": {"positive": ["id1", "id2"], "negative": ["id3"]}}; 3. discover: Explore the vector space by moving away from certain points. Example: {"discover": {"target": "id1", "context": [{"positive": "id2"}]}}; 4. context: Search with contextual constraints. Example: {"context": [{"positive": "id1", "negative": "id2"}]}; 5. order_by: Retrieve points sorted by a payload field (non-vector search). Example: {"order_by": {"key": "timestamp", "direction": "desc"}}. Note: Vector dimension must match the collection's configuration.

Options:

• FilterCondition: Filtering criteria using Qdrant Filter DSL. Supports boolean logic: 1. must: All conditions must match (AND); 2. should: At least one condition must match (OR); 3. must_not: None should match (NOT). Supported Operators: match (Exact match, e.g., {"key": "city", "match": {"value": "London"}}), match any (Match any value, e.g., {"key": "tag", "match": {"any": ["tech", "ai"]}}), range (Numeric/date range, e.g., {"key": "age", "range": {"gt": 25}}), geo_radius (Geographic proximity), has_id (Match specific IDs). Example: {"must": [{"key": "category", "match": {"value": "AI"}}], "should": [{"key": "priority", "match": {"any": ["high"]}}]}
• Limit: Maximum number of results to return. Default: 10
• Offset: Number of results to skip for pagination. Default: 0
• WithPayload: Whether to include 'payload' (metadata/attributes associated with the point, such as name, category, or description) in the results. Default: true.
• WithVectors: Whether to include 'vectors' (the raw numerical embedding arrays used for similarity calculation) in the results. Usually set to false unless you need to perform further local processing. Default: false.
• Options: Advanced search parameters. Supports: 1. Params (object): HNSW (Hierarchical Navigable Small World - a fast vector indexing algorithm) tuning. hnsw_ef (int, 4-512): Search accuracy. Higher = more accurate but slower. Default: 16. exact (bool): Force brute-force search (bypasses HNSW index) for highest accuracy. Default: false. 2. ScoreThreshold (float, 0.0-1.0): Minimum similarity score filter. Only return results with score ≥ this value. 3. Consistency (string): Read consistency level. Default: 'eventual'. 'eventual': Fastest, may read stale data; 'majority': Balanced, reads from majority replicas; 'quorum': Strongest consistency, slowest.

Output:

• Result (bool): Boolean indicating whether the vector query operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where vector queries were executed.
• QueryType (string): Type of query executed (nearest, recommend, discover, context, order_by, etc.).
• Results (object-array): List of found points. Each item contains: id (string/int): Unique identifier; score (float): Similarity score (0.0 to 1.0); payload (object): Metadata pairs; vector (number-array or null): The raw embedding vector. If WithVectors=true, returns the array; if WithVectors=false, returns null. Returns empty array [] if no matches found.
• ResultsCount (number): Number of results returned by the vector query operation.
• SearchParams (object): The actual parameters applied by the search engine. Includes: limit (int): Max results; offset (int): Pagination start; with_payload (bool): Metadata status; with_vectors (bool): Vector status; hnsw_ef (int): Accuracy setting; exact (bool): If brute-force was used; score_threshold (float): Minimum similarity filter; consistency (string): Consistency level used.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description if operation fails, empty string on success.

Find Points Groups

Execute grouped vector queries in a Qdrant collection. Use this to group search results by a specific payload field (e.g., category or user_id); for standard searches use 'Find Points'.

Required: CollectionName - target collection; Query - search config (supports Nearest, Recommend, Discover, Context, Order By, Fusion, Sample, Formula); GroupBy - payload field to group by.

Optional: GroupSize (default: 3) - points per group; Limit (default: 10) - max groups; FilterCondition - boolean filter logic (must/should/must_not); Consistency (values: eventual/majority/quorum, default: server setting); AdditionalParams - extended settings (hnsw_ef, exact, with_payload, etc.).

Returns: Groups (object-array) - grouped results. Key fields: id (group value), hits (array of points with id, score, payload, vector). Plus Result, Status, Time, CollectionName, QueryType, GroupBy, GroupsCount, TotalHits, HasMore, SearchParams, StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Groups results by the specified field value.

Limitations: 404 if collection not found; 400 if query invalid.

Input Parameters:

• CollectionName: Name of the Qdrant collection to search within. Example: "my_documents"
• Query: The search query defining how to find and rank points. Supports 8 specialized types: 1. Nearest: Semantic search for most similar points. Example: {"nearest": [0.1, 0.2, 0.3]} 2. Recommend: Personalized recommendation based on positive/negative examples. Example: {"recommend": {"positive": [101, 102], "negative": [103]}} 3. Discover: Exploratory search to find points similar to target but in a different context. Example: {"discover": {"target": 101, "context": [{"positive": 102, "negative": 103}]}} 4. Context: Refine search results within a specific set of positive/negative constraints. Example: {"context": [{"positive": 101, "negative": 102}]} 5. Order By: Rank points by a specific metadata field (e.g., price, time). Example: {"order_by": {"key": "timestamp", "direction": "desc"}} 6. Fusion: Combine results from multiple queries using algorithms like RRF. Example: {"fusion": "rrf"} 7. Sample: Retrieve a random subset of points for diversity. Example: {"sample": "random"} 8. Formula: Use a custom mathematical expression for scoring. Example: {"formula": "log(1 + count) * score"}
• GroupBy: The name of the payload field to group results by. Each group will contain points with the same value in this field. Example: "category"

Options:

• GroupSize: Maximum number of points to return per group. Default: 3. Example: 5
• Limit: Maximum number of groups to return in the result. Default: 10. Example: 5
• FilterCondition: Boolean filter logic to restrict search results. - must: All conditions MUST be met (SQL AND). Example: {"must": [{"key": "status", "match": {"value": "active"}}]} - should: At least one condition should be met (SQL OR). Example: {"should": [{"key": "color", "match": {"value": "red"}}, {"key": "color", "match": {"value": "blue"}}]} - must_not: Exclude points matching these conditions (SQL NOT). Example: {"must_not": [{"key": "category", "match": {"value": "trash"}}]}
• Consistency: Data freshness level (Read Consistency): - 'eventual': Fastest, but might return slightly outdated data. - 'majority': Balanced, ensures data is seen by most nodes before reading. - 'quorum': Strictest, ensures you read the most recent update across the cluster. Default is server-side setting.
• AdditionalParams: Extended search configuration: - params: Technical tuning. 'hnsw_ef' (1-512) controls search effort in the HNSW graph; higher means more accuracy but slower speed. Default 128. - with_payload: Whether to return metadata. (boolean) - with_vectors: Whether to return vector data. (boolean) - score_threshold: Minimum score to include a point. (float) - with_lookup: Fetch extra data from another collection. Example: {"with_lookup": {"collection": "users", "with_payload": true}}

Output:

• Result (bool): Boolean indicating whether the grouped query operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where grouped queries were executed.
• QueryType (string): Type of query executed (nearest, recommend, discover, context, etc.).
• GroupBy (string): Payload field name used for grouping the query results.
• Groups (object-array): Array of grouped results. Each group contains: - hits: Array of point objects. Each point includes 'id' (string/int), 'score' (float), 'payload' (JSON object containing metadata like name, category, etc.), and 'vector' (array of floats representing the embedding). - id: The value of the field used for grouping (string/int/null).
• GroupsCount (number): Number of groups returned in this response.
• TotalHits (number): Total number of points found across all returned groups.
• HasMore (bool): Indicates whether there are more groups beyond the current limit. True if the number of returned groups equals the limit parameter.
• SearchParams (object): The actual parameters applied during search: - hnsw_ef: Search effort level used. - exact: Whether exact search was performed. - quantization: Object describing compression settings used, including 'type' (e.g., scalar, product), 'rescore' (boolean), and 'oversampling' (float).
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Retrieve Point

Retrieve a specific point from a Qdrant collection by ID. Use this to get full vector data and payload; for searching use 'Search Points'.

Required: CollectionName - target collection; PointId - integer or UUID string.

Optional: Consistency (values: eventual/majority/quorum, default: server default).

Returns: PointData (object) - complete point info including id, vector, payload. Key fields: id (str/int), vector (list), payload (object). Plus Status, Time, StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: PointId from 'Search Points' or 'Scroll Points' action.

Limitations: Returns 404 if point not found.

Input Parameters:

• CollectionName: The name of the collection where the point is stored. Example: "my_collection".
• PointId: The unique identifier of the point to retrieve (supports integer or UUID string). You can obtain this ID from the results of 'Search Points' or 'Scroll Points' actions. Example: "550e8400-e29b-41d4-a935-446655440000".

Options:

• Consistency: Read consistency level for distributed environments. Values: 'eventual' (fastest), 'majority' (balanced), 'quorum' (most accurate). Default: '' (server default).

Output:

• PointData (object): Complete point information including: id (string/number), vector (array/object), payload (object), shard_key (string).
• Result (bool): Boolean indicating whether the point retrieval operation was successful.
• Status (string): Operation status information returned by Qdrant API (e.g., 'ok').
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name from which the point was retrieved.
• PointId (string): Point ID that was queried for retrieval.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Retrieve Points

Retrieve multiple points from a Qdrant collection by IDs. Use this to get details of specific points; for searching use 'Search Points'.

Required: CollectionName - target collection name; Ids - list of point IDs (strings or numbers).

Optional: WithPayload - return metadata (default: true); WithVectors - return vectors (default: false); Consistency - read consistency level (values: eventual/majority/quorum, default: empty).

Returns: PointsData (object-array) - list of retrieved points. Key fields: id (string/number), payload (object), vector (array), shard_key (string). Plus FoundCount, RequestedIds, StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: Ids from 'Search Points' or 'Scroll Points'.

Behavior: Returns found points; missing IDs are ignored (not returned). Important: Non-existent IDs are silently ignored without errors. If you request 10 IDs but receive only 8 points, it means 2 IDs do not exist. Use FoundCount vs RequestedIds length to detect missing records.

Limitations: 404 if collection not found; 400 if IDs format is invalid.

Input Parameters:

• CollectionName: The name of the collection to retrieve points from (required).
• Ids: List of point IDs to retrieve. Supports both string (UUID) and numeric IDs. IDs can be obtained from: 1. 'Search Points' action (returns matching IDs), 2. 'Scroll Points' action (for batch collection), 3. External databases. Example: ["doc_001", "doc_002"] or [101, 102].

Options:

• WithPayload: Whether to return business metadata (payload). Set to true to see fields like 'title' or 'content'. Default: true.
• WithVectors: Whether to return raw vector embeddings (numerical arrays). Default: false. Set to true only when you need to calculate vector similarity manually, debug embedding quality, or export vectors for external processing. For normal metadata retrieval, keep this false to reduce response size.
• Consistency: Read consistency level for distributed clusters. Values: 'eventual' (fastest), 'majority' (balanced), 'quorum' (strongest). If you are unsure whether your Qdrant is a distributed cluster, leave this empty to use the default single-node behavior.

Output:

• Result (bool): Boolean indicating whether the batch points retrieval operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name from which points were retrieved.
• RequestedIds (string-array): The original list of IDs from the request. Type matches input: string array for string IDs, number array for numeric IDs. Example: ["doc_001", "doc_002"] or [101, 102].
• PointsData (object-array): List of retrieved points. Each object contains: 'id' (string|number): unique identifier; 'payload' (object|null): business metadata (e.g., {"title": "...", "author": "..."}), null if WithPayload=false. Note: The payload structure is completely flexible and depends on the collection schema; 'vector' (number-array|null): vector embedding data, null if WithVectors=false; 'shard_key' (string|null): shard identifier for distributed deployments.
• FoundCount (number): Number of points successfully found and retrieved from the collection.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found).
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Scroll Points

Scroll through points in a Qdrant collection. Use this to retrieve points with pagination and filtering; for searching by vector similarity use 'Search Points'.

Required: CollectionName - target collection name.

Optional: FilterCondition - JSON object for filtering (supports must/should/range); Limit - points per page (default: 10); Offset - pagination anchor (UUID or int); OrderBy - sorting config; WithPayload - include metadata (default: true); WithVectors - include vectors (default: false); Consistency - read consistency level.

Returns: Points (object-array) - list of retrieved points. Key fields: id (string/number), payload (object), vector (array/object). Plus Result, Status, Time, NextPageOffset, HasNextPage, ReturnedCount, FilterApplied, OriginalStatusCode, StatusCode, ErrorMessage.

Behavior: Returns points in defined order. Limit max 10000.

Limitations: 404 if collection not found; 400 if filter syntax invalid.

Input Parameters:

• CollectionName: Name of the collection to scroll points from. Must consist of alphanumeric characters, hyphens, or underscores. Example: "my_collection_v1"

Options:

• FilterCondition: Filter condition object using Qdrant syntax. Supports: 'must' (All conditions must match - AND logic), 'should' (At least one condition must match - OR logic), 'must_not' (None of the conditions should match - NOT logic). Example: {"must": [{"key": "status", "match": {"value": "active"}}]}
• Limit: Number of points to return per page. Range: 1-10000. Values above 10000 will be truncated by the server. Default: 10.
• Offset: Pagination anchor. Can be a Point ID (UUID string or integer) from 'NextPageOffset' of a previous response, or a numeric string representing the number of points to skip. Leave empty for the first page. Example: "550e8400-e29b-41d4-a716-446655440000"
• OrderBy: Sorting configuration. Supports single-field sorting. Format: {'key': 'field_name', 'direction': 'asc'/'desc'}. Example: {"key": "created_at", "direction": "desc"}
• WithPayload: Whether to include payload data (business metadata fields like title, tags, etc.) in the response. Default: true.
• WithVectors: Whether to include vector embeddings (AI-generated numeric arrays used for similarity search) in the response. Note: Enabling this significantly increases response size. Default: false. Example: true
• Consistency: Read consistency level. Controls data freshness vs performance trade-off: 'eventual' (Fastest, may read slightly outdated data), 'majority' (Balanced, reads from most nodes), 'quorum' (Strongest, ensures latest data). Default: empty (server default). Example: "eventual"

Output:

• Result (bool): Boolean indicating whether the scroll operation was successful.
• Status (string): Operation status information returned by Qdrant API.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where scrolling was performed.
• Points (object-array): Array of retrieved point objects. Each object contains: id (string|number): Unique point identifier; payload (object): Business metadata, commonly includes fields like title, name, tags, or timestamps; vector (array|object): Vector embeddings (only present if WithVectors=true); order_value (any): Sorting value (only present if OrderBy is specified).
• NextPageOffset (string): Offset for the next page (empty if no more pages available).
• CurrentLimit (number): Current page size limit used for this scroll operation.
• HasNextPage (bool): Boolean indicating whether there are more pages available for scrolling.
• ReturnedCount (number): Number of points actually returned in the current page.
• FilterApplied (bool): Boolean indicating whether a filter condition was applied during scrolling.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Upsert Points

Upsert points into a Qdrant collection. Use this to insert new points or update existing ones by ID; for searching use 'Search Points'.

Required: CollectionName - target collection name; Points - list of point objects (id, vector, payload).

Optional: Wait (default: true) - wait for indexing; Ordering (values: weak/medium/strong, default: weak) - consistency level.

Returns: Result (bool), Status (string), Time (number), CollectionName (string), PointsCount (number), OperationId (number). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: CollectionName from 'List Collections' action.

Behavior: Overwrites existing points with the same ID. If Wait is true, returns after indexing is complete.

Limitations: 404 if collection does not exist. 400 if vector dimension mismatches.

Input Parameters:

• CollectionName: The name of the target collection to upsert points into. You can obtain existing collection names by calling the 'List Collections' action or via the Qdrant Dashboard. Format: alphanumeric, underscores, or hyphens, 1-255 characters. Example: "my_vectors"
• Points: An array of point objects to insert or update. Each object must include: 'id' (integer or string UUID), 'vector' (list of floats representing the embedding), and optionally 'payload' (object containing metadata). Example: [{"id": 1, "vector": [0.1, 0.2, 0.3], "payload": {"category": "electronics"}}]

Options:

• Wait: Defines the synchronization mode. If 'true' (recommended), the action waits for the data to be fully written and indexed before returning, ensuring immediate searchability. If 'false', the action returns an OperationId immediately for asynchronous processing. Example: true
• Ordering: Specifies the write consistency level. 'weak' (default) offers the best performance with eventual consistency; 'medium' balances speed and visibility; 'strong' ensures data is immediately visible across all nodes but is slower. Use 'weak' for bulk ingestion and 'strong' for critical real-time updates. Example: "weak"

Output:

• Result (bool): Boolean indicating whether the upsert operation was successful.
• Status (string): The status of the upsert operation. Possible values: "ok" (operation successful), "error" (operation failed, check ErrorMessage for details). Example: "ok"
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where points were upserted.
• PointsCount (number): Number of points that were successfully inserted or updated.
• OperationId (number): Operation ID for tracking asynchronous operations (0 if synchronous).
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors), 500 = system error, -1 = parameter validation error.
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully.

Delete Vectors

Delete specific named vectors from points in a Qdrant collection. Use this to remove specific vector fields (e.g., 'text_vector') from points while preserving the points and other data; to delete points entirely use 'Delete Points'.

Required: CollectionName - target collection name; Vectors - list of vector names to delete.

Optional: Points - list of point IDs (integers or UUID strings); FilterCondition - filter criteria to select points; Wait (default: true) - wait for disk persistence; Ordering (values: weak/medium/strong, default: weak).

Returns: Result (bool), Status (string), Time (number), CollectionName (string), VectorNames (string-array), TargetType (string), OperationId (number). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Behavior: Must provide either 'Points' or 'FilterCondition'.

Limitations: 404 if collection missing.

Input Parameters:

• CollectionName: The name of the collection where vectors will be deleted. You can retrieve available collection names using the 'List Collections' action. Example: "my_knowledge_base"
• Vectors: List of vector names to delete from the points. To find available vector names, use the 'Get Collection Info' action. Common names include 'text', 'image', or 'dense'. Example: ["text_vector", "image_vector"]

Options:

• Points: List of point IDs to delete vectors from. IDs can be integers (will be auto-converted) or UUID strings. IMPORTANT: Either 'Points' or 'FilterCondition' must be provided. To find point IDs, use 'Search Points' or 'Scroll Points' actions. Example: ["1", "2", "550e8400-e29b-41d4-a716-446655440000"]
• FilterCondition: Filter condition to select points for vector deletion. Supports Qdrant filter syntax. IMPORTANT: Either 'Points' or 'FilterCondition' must be provided. Structure: - 'must': AND logic array - 'should': OR logic array - 'must_not': NOT logic array Example (Match): {"must": [{"key": "category", "match": {"value": "outdated"}}]} Example (Range): {"must": [{"key": "price", "range": {"lt": 100}}]}
• Wait: Whether to wait for the deletion to be physically committed to the disk before returning. Set to 'true' for immediate consistency (recommended for small updates), or 'false' for asynchronous background processing (recommended for large batches).
• Ordering: Write consistency guarantee level. Defines how the operation is replicated across nodes. Options: 'weak' (default): Fastest. Use for non-critical batch operations where eventual consistency is acceptable. 'medium': Balanced. Use for standard production operations. 'strong': Slowest but highest consistency. Use for critical operations where data loss is unacceptable (e.g., financial records). Recommendation: Leave empty to use default 'weak' for most cases.

Output:

• Result (bool): Boolean indicating whether the vector deletion was successful.
• Status (string): The status of the deletion operation. Possible values: 'ok': Operation successfully completed (when Wait=true); 'pending': Operation is queued for async processing (when Wait=false); 'error': Operation failed (check ErrorMessage for details). Note: When Wait=true (default), you will always get 'ok' or an error. When Wait=false, you get 'pending' and should track via OperationId.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where vectors were deleted.
• VectorNames (string-array): List of vector names that were deleted from the points.
• TargetType (string): The method used to identify points for deletion. Values: 'points' (explicit IDs used), 'filter' (criteria used), 'mixed' (both IDs and criteria used).
• OperationId (number): Operation ID for tracking the vector deletion process.
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

Update Vectors

Update vector data for specified points in a Qdrant collection. Use this to modify vectors for existing points without affecting payload; for creating new points use 'Upsert Points'.

Required: CollectionName - target collection name; Points - list of point objects containing 'id' and 'vector'.

Optional: Wait (default: true) - wait for operation completion; Ordering (values: weak/medium/strong, default: weak) - write consistency level.

Returns: Result (bool), Status (string), Time (number), CollectionName (string), UpdatedCount (number), OperationId (number). Plus StatusCode, ErrorMessage, OriginalStatusCode.

Dependency: CollectionName from 'List Collections' or 'Create Collection'.

Behavior: Updates vectors for existing points. If a point does not exist, it is typically skipped or handled according to Qdrant configuration.

Limitations: 404 if collection does not exist. 400 if vector dimensions do not match collection configuration.

Input Parameters:

• CollectionName: Target collection name.
• Points: A list of point objects to update. Each object must contain an 'id' and the new 'vector' data. Structure per Point: - id (integer or string): Unique identifier of the point to update. - vector (array or object): The new vector values. - For single vector collections: [0.1, 0.2, ...] - For named vector collections: {"vector_name": [0.1, 0.2, ...]} Example Template: [ { "id": 1, "vector": [0.05, 0.61, 0.76] }, { "id": "point-002", "vector": { "text_vector": [0.1, 0.2, 0.3], "image_vector": [0.4, 0.5, 0.6] } } ]

Options:

• Wait: Determines if the request should block until the update is physically committed to the disk and indexed. - true (Default): Synchronous mode. The response is returned only after the update is finished. Recommended for small batches to ensure data is searchable immediately. - false: Asynchronous mode. The API returns immediately with an OperationId. Recommended for large background updates to avoid timeouts.
• Ordering: Defines the write consistency guarantees for the operation. This determines how many replicas must acknowledge the update before returning. - weak: (Default) Only one replica needs to acknowledge. Highest performance but lower consistency. - medium: Multiple replicas acknowledge. Balanced performance and consistency. - strong: All replicas must acknowledge. Ensures immediate consistency across the cluster but has higher latency.

Output:

• Result (bool): Boolean indicating whether the vector update operation was successful.
• Status (string): The operational status returned by Qdrant. Possible values are: - "ok": The update was accepted and processed successfully. - "error": The update failed. Check ErrorMessage for details.
• Time (number): Operation execution time in seconds.
• CollectionName (string): Target collection name where vectors were updated.
• UpdatedCount (number): Number of points that were successfully updated with new vector data.
• OperationId (number): Operation ID for tracking asynchronous operations (0 if synchronous).
• OriginalStatusCode (number): The raw HTTP status code returned by the Qdrant API (e.g., 200 for success, 404 for not found). Default 0 means no HTTP request was made.
• StatusCode (number): System operation status code. 200 = system healthy (including API business errors reflected in ErrorMessage), 500 = system error (network failure, timeout, code crash), -1 = parameter validation error (request not sent).
• ErrorMessage (string): Error description message. Empty string if the operation completed successfully. Contains specific error details when StatusCode is non-200 or when the upstream API returns a business error.

5. Example Usage

This section will guide you through creating a simple workflow to create a new vector collection in your Qdrant database.

The workflow consists of: Start -> Qdrant (Create Collection) -> Answer.

1. Add the Tool Node:
   • On the workflow canvas, click the "+" icon to add a new node.
   • Select the "Tools" tab in the panel.
   • Find and select Qdrant from the tool list.
   • Choose the Create Collection action from the supported operations list to add the node to the canvas.
2. Configure the Node:
   • Click on the newly added Create Collection node to open its configuration panel on the right.
   • Credentials: At the top of the panel, select your configured Qdrant credentials from the dropdown menu.
   • CollectionName: Enter a unique name for your new collection, for example, "my_first_collection".
   • VectorsConfig: Provide the vector configuration JSON, specifying the size and distance metric. For example: {"size": 1536, "distance": "Cosine"}.
3. Run and Verify:
   • Once all required parameters are filled, the error indicator on the node will disappear.
   • Click the "Test Run" button at the top right of the canvas to execute the workflow.
   • After a successful run, you can click the log icon to view the detailed input and output, verifying that the collection was created successfully.

After completing these steps, your workflow is fully configured. Upon execution, a new vector collection will be initialized in your Qdrant database.

6. FAQs

Q: Why am I getting a 401 Unauthorized error?

A: Please check the following points:

• API Key: Ensure that the API Key provided in your credentials is correct and has not expired.
• Cluster URL: Verify that the Qdrant cluster URL is accurate and accessible.

Q: Why does the 'Create Collection' action return a 409 error?

A: This indicates a conflict:

• Existing Collection: A collection with the specified CollectionName already exists in your Qdrant instance. Please use a different name or delete the existing one first.

7. Official Documentation

Qdrant Official API Documentation