Index Management

The vector database supports different types of indexes for optimizing vector search operations. Each collection can have separate indexes for dense vectors, sparse vectors, and text search.

Create Dense Vector Index

Creates an index for dense vector search using approximate nearest neighbor (ANN) algorithms.

Endpoint: POST /vectordb/collections/{collection_id}/indexes/dense

URL Parameters:

Parameter	Type	Required	Description
collection_id	string	Yes	ID (name) of the collection to create an index for.

Request Body:

{
  "name": "index_name",
  "distance_metric_type": "cosine",
  "quantization": {
    "type": "auto",
    "properties": {
      "sample_threshold": 100
    }
  },
  "index": {
    "type": "hnsw",
    "properties": {
      "num_layers": 7,
      "max_cache_size": 1000,
      "ef_construction": 512,
      "ef_search": 256,
      "neighbors_count": 32,
      "level_0_neighbors_count": 64
    }
  }
}

Request Parameters:

Parameter	Type	Required	Description
name	string	Yes	Name of the index.
distance_metric_type	string	Yes	Distance metric to use: “cosine”, “euclidean”, or “dot_product”.
quantization	object	Yes	Configuration for vector quantization.
quantization.type	string	Yes	Type of quantization: “auto” or “scalar”.
quantization.properties	object	Yes	Properties specific to the quantization type.
quantization.properties.sample_threshold	integer	Yes (if auto)	Number of vectors to sample for automatic quantization.
quantization.properties.data_type	string	Yes (if scalar)	Data type for scalar quantization: “binary”, “quaternay”, “octal”, “u8”, “f16”, or “f32”.
quantization.properties.range	object	Yes (if scalar)	Value range for scalar quantization.
index	object	Yes	Configuration for index algorithm.
index.type	string	Yes	Type of index algorithm. Currently only “hnsw” is supported.
index.properties	object	Yes	Properties specific to the index algorithm.
index.properties.num_layers	integer	No	Number of layers in the HNSW graph.
index.properties.max_cache_size	integer	No	Maximum number of elements in the cache.
index.properties.ef_construction	integer	No	Size of the dynamic candidate list during index construction.
index.properties.ef_search	integer	No	Size of the dynamic candidate list during search.
index.properties.neighbors_count	integer	No	Number of established connections per node.
index.properties.level_0_neighbors_count	integer	No	Number of connections at the ground layer.

Response:

{}

Status Codes:

Code	Description
201	Created. Index created successfully.
400	Bad Request. Invalid index parameters.
404	Not Found. Collection not found.
409	Conflict. Index already exists for this collection.
500	Server Error. Failed to create index.

Create Sparse Vector Index

Creates an index for sparse vector search.

Endpoint: POST /vectordb/collections/{collection_id}/indexes/sparse

URL Parameters:

Parameter	Type	Required	Description
collection_id	string	Yes	ID (name) of the collection to create an index for.

Request Body:

{
  "name": "index_name",
  "quantization": 64,
  "sample_threshold": 1000
}

Request Parameters:

Parameter	Type	Required	Description
name	string	Yes	Name of the index.
quantization	integer	Yes	Quantization bit value. Allowed values: 16, 32, 64, 128, or 256.
sample_threshold	integer	Yes	Number of vectors to sample for calibrating the index.

Response:

{}

Status Codes:

Code	Description
201	Created. Index created successfully.
400	Bad Request. Invalid index parameters.
404	Not Found. Collection not found.
409	Conflict. Index already exists for this collection.
500	Server Error. Failed to create index.

Create TF-IDF Index

Creates an index for text search using TF-IDF/BM25.

Endpoint: POST /vectordb/collections/{collection_id}/indexes/tf-idf

URL Parameters:

Parameter	Type	Required	Description
collection_id	string	Yes	ID (name) of the collection to create an index for.

Request Body:

{
  "name": "index_name",
  "sample_threshold": 1000,
  "k1": 1.2,
  "b": 0.75
}

Request Parameters:

Parameter	Type	Required	Description
name	string	Yes	Name of the index.
sample_threshold	integer	Yes	Number of documents to sample for calibrating the index.
k1	float	Yes	BM25 k1 parameter that controls term frequency saturation. Typical values range from 1.2 to 2.0.
b	float	Yes	BM25 b parameter that controls document length normalization. Typical value is 0.75.

Response:

{}

Status Codes:

Code	Description
201	Created. Index created successfully.
400	Bad Request. Invalid index parameters.
404	Not Found. Collection not found.
409	Conflict. Index already exists for this collection.
500	Server Error. Failed to create index.

Get Index

Retrieves information about the indexes defined for a collection.

Endpoint: GET /vectordb/collections/{collection_id}/indexes

URL Parameters:

Parameter	Type	Required	Description
collection_id	string	Yes	ID (name) of the collection to get indexes for.

Response:

{
  "dense": {
    "name": "dense_index_name",
    "distance_metric_type": "cosine",
    "quantization": {
      "type": "auto",
      "properties": {
        "sample_threshold": 100
      }
    },
    "index": {
      "type": "hnsw",
      "properties": {
        "num_layers": 7,
        "max_cache_size": 1000,
        "ef_construction": 512,
        "ef_search": 256,
        "neighbors_count": 32,
        "level_0_neighbors_count": 64
      }
    }
  },
  "sparse": {
    "name": "sparse_index_name",
    "quantization": 64,
    "sample_threshold": 1000
  },
  "tf-idf": {
    "name": "tf_idf_index_name",
    "sample_threshold": 1000,
    "k1": 1.2,
    "b": 0.75
  }
}

Status Codes:

Code	Description
200	Success. Index information returned.
404	Not Found. Collection not found.
500	Server Error. Failed to retrieve index information.

Delete Index

Deletes an index from a collection.

Endpoint: DELETE /vectordb/collections/{collection_id}/indexes/{index_type}

URL Parameters:

Parameter	Type	Required	Description
collection_id	string	Yes	ID (name) of the collection containing the index.
index_type	string	Yes	Type of index to delete: “dense”, “sparse”, or “tf_idf”.

Status Codes:

Code	Description
204	No Content. Index deleted successfully.
400	Bad Request. Invalid index type.
404	Not Found. Collection or index not found.
500	Server Error. Failed to delete index.