Syntax

{
  "createIndexes": "<collection_name>",
  "indexes": [
    { "name": "<index_name>",
      "key": { "<path_to_property>": "vector" },
      "kaiIndexOptions": {
        "index_type": "string"
        "dimensions": <integer_value>,
        // other fields relevant to each index type }
    } ] }

Remarks

• The createIndexes command creates an index on the specified field in a collection using a persisted computed column within the same collection.

• SingleStore recommends creating collections (or tables) before creating a vector index. You must specify the field that holds the vector embeddings while creating the collection. Here's a sample syntax:

  MongoDB

  Copy

  db.createCollection(
    "<collection_name>",
    columns= [{'id': '<vector_field_name>', 'type': "VECTOR(<vector_dimension>, <type>) NOT NULL"}]
  )

  Refer to createCollection for more information.

• If the specified collection does not exist, it is created along with the required structures. For example:

  MongoDB

  Copy

  db.runCommand({
    createIndexes: "sampleCollection", 
    indexes: [
       {  key: { "sampleCollection.vectorField": "vector" }, 
       name: "searchVector_IVF_EUC", 
       kaiIndexOptions: 
         { "index_type":"IVF_FLAT", 
           "nlist":5, 
           "nprobe":1, 
           "metric_type": "EUCLIDEAN_DISTANCE", 
           "dimensions": 2 } }
    ]
  } )

  SQL

  Copy

  DESC sampleCollection;

  +-------------------------------+----------------+------+-----+---------+----------+
  | Field                         | Type           | Null | Key | Default | Extra    |
  +-------------------------------+----------------+------+-----+---------+----------+
  | _id                           | bson           | NO   |     | NULL    |          |
  | _more                         | bson           | NO   |     | NULL    |          |
  | $_id                          | longblob       | YES  | UNI | NULL    | computed |
  | $sampleCollection.vectorField | vector(2, F32) | NO   | MUL | NULL    | computed |
  +-------------------------------+----------------+------+-----+---------+----------+

• Use the index_type and dimensions fields in the kaiIndexOptions parameter to specify the type of vector index and the dimension of the vector, respectively.

• You can specify supported Index Options in the kaiIndexOptions parameter for each index type. The specified values are set as defaults for the respective index search parameters.

Example

The following example shows how to create a vector index using the createIndexes command.

db.runCommand({
  createIndexes: "exampleCollection", 
  indexes: [
     {  key: { "exampleCollection.vectorField": "vector" }, 
     name: "searchVector_IVF_EUC", 
     kaiIndexOptions: 
       { "index_type":"IVF_FLAT", 
         "nlist":5, 
         "nprobe":1, 
         "metric_type": "EUCLIDEAN_DISTANCE", 
         "dimensions": 2 } }
  ]
} )

Use the db.<collection_name>.getIndexes() command to view existing indexes on a collection. For example:

db.exampleCollection.getIndexes()

[ { v: 2, key: { _id: 1 }, name: '_id_' },
  { v: 2,
    key: { 'exampleCollection.vectorField': 'vector' },
    name: 'searchVector_IVF_EUC',
    kaiSearchSpec: 
     { nlist: 5,
       nprobe: 1,
       index_type: 'IVF_FLAT',
       metric_type: 'EUCLIDEAN_DISTANCE',
       dimensions: 2 } } ]