Using the Momento Vector Index API

Momento Vector Index (MVI) is a scalable, developer-friendly vector index service designed for real-time storage and retrieval of vector data for use in AI-powered applications.

Vector Index Client

To interact with Momento Vector Indexes, you must use a VectorIndexClient.

JavaScript

new PreviewVectorIndexClient({
  credentialProvider: CredentialProvider.fromEnvironmentVariable({
    environmentVariableName: 'MOMENTO_API_KEY',
  }),
  configuration: VectorIndexConfigurations.Laptop.latest(),
});

Python

PreviewVectorIndexClientAsync(
    VectorIndexConfigurations.Default.latest(), CredentialProvider.from_environment_variable("MOMENTO_API_KEY")
)

Vector Index methods

Create Index

Creates a vector index.

Name	Type	Description
indexName	String	Name of the vector index.
numDimensions	String	Number of dimensions per vector.
similarityMetric	String	Metric used to quantify the distance between vectors. Can be cosine similarity, inner product, or euclidean similarity. Defaults to cosine similarity.

Method response object

• Success
• AlreadyExists
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.createIndex('test-index', 2);
if (result instanceof CreateVectorIndex.Success) {
  console.log("Index 'test-index' created");
} else if (result instanceof CreateVectorIndex.AlreadyExists) {
  console.log("Index 'test-index' already exists");
} else if (result instanceof CreateVectorIndex.Error) {
  throw new Error(
    `An error occurred while attempting to create index 'test-index': ${result.errorCode()}: ${result.toString()}`
  );
}

Python

response = await vector_client.create_index("test-index", 2)
match response:
    case CreateIndex.Success():
        print("Index 'test-index' created")
    case CreateIndex.IndexAlreadyExists():
        print("Index 'test-index' already exists")
    case CreateIndex.Error() as error:
        print(f"Error creating index 'test-index': {error.message}")

Delete Index

Deletes a vector index.

Name	Type	Description
indexName	String	Name of the vector index.

Method response object

• Success
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.deleteIndex('test-index');
if (result instanceof DeleteVectorIndex.Success) {
  console.log("Index 'test-index' deleted");
} else if (result instanceof DeleteVectorIndex.Error) {
  throw new Error(
    `An error occurred while attempting to delete index 'test-index': ${result.errorCode()}: ${result.toString()}`
  );
}

Python

response = await vector_client.delete_index("test-index")
match response:
    case DeleteIndex.Success():
        print("Index 'test-index' deleted")
    case DeleteIndex.Error() as error:
        print(f"Error deleting index 'test-index': {error.message}")

List Indexes

Lists all vector indexes.

Method response object

• Success
  • getIndexes(): VectorIndexInfo[]
    • VectorIndexInfo:
      • name: String
      • numDimensions: number
      • similarityMetric: String
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.listIndexes();
if (result instanceof ListVectorIndexes.Success) {
  console.log(
    `Indexes:\n${result
      .getIndexes()
      .map(index => index.toString())
      .join('\n')}\n\n`
  );
} else if (result instanceof ListVectorIndexes.Error) {
  throw new Error(`An error occurred while attempting to list caches: ${result.errorCode()}: ${result.toString()}`);
}

Python

response = await vector_client.list_indexes()
match response:
    case ListIndexes.Success() as success:
        print(f"Indexes:\n{success.indexes}")
    case CreateIndex.Error() as error:
        print(f"Error listing indexes: {error.message}")

Upsert Item Batch

Upserts a batch of items into a vector index.

Name	Type	Description
indexName	String	Name of the vector index.
items	VectorIndexItem	Items to upsert into the index.

Method response object

• Success
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.upsertItemBatch('test-index', [
  {
    id: 'example_item_1',
    vector: [1.0, 2.0],
    metadata: {key1: 'value1'},
  },
  {
    id: 'example_item_2',
    vector: [3.0, 4.0],
    metadata: {key2: 'value2'},
  },
]);
if (result instanceof VectorUpsertItemBatch.Success) {
  console.log('Successfully added items');
} else if (result instanceof VectorUpsertItemBatch.Error) {
  throw new Error(`An error occurred while adding items to index: ${result.errorCode()}: ${result.toString()}`);
}

Python

response = await vector_client.upsert_item_batch(
    "test-index",
    [
        Item(id="example_item_1", vector=[1.0, 2.0], metadata={"key1": "value1"}),
        Item(id="example_item_2", vector=[3.0, 4.0], metadata={"key2": "value2"}),
    ],
)
match response:
    case UpsertItemBatch.Success():
        print("Successfully added items to index 'test-index'")
    case UpsertItemBatch.Error() as error:
        print(f"Error adding items to index 'test-index': {error.message}")

Search

Searches for items with vectors most similar to a query vector.

Name	Type	Description
indexName	String	Name of the vector index.
queryVector	number[]	Vector to search for.
topK	number	Number of results to return. Defaults to 10.
metadataFields	String[] or ALL_VECTOR_METADATA	A list of metadata fields to return with each result, or a value indicating all metadata should be returned. If not provided, no metadata is returned. Defaults to None.

Method response object

• Success
  • hits(): SearchHit[]
    • SearchHit:
      • id: string
      • score: number
      • metadata: Map<string, string>
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.search('test-index', [1.0, 2.0], {topK: 3, metadataFields: ALL_VECTOR_METADATA});
if (result instanceof VectorSearch.Success) {
  console.log(`Found ${result.hits().length} matches`);
} else if (result instanceof VectorSearch.Error) {
  throw new Error(`An error occurred searching index test-index: ${result.errorCode()}: ${result.toString()}`);
}

Python

response = await vector_client.search("test-index", [1.0, 2.0], top_k=3, metadata_fields=ALL_METADATA)
match response:
    case Search.Success() as success:
        print(f"Found {len(success.hits)} matches")
    case Search.Error() as error:
        print(f"Error searching index 'test-index': {error.message}")

Search And Fetch Vectors

Searches for items with vectors most similar to a query vector. Also returns corresponding vectors to search hits.

Name	Type	Description
indexName	String	Name of the vector index.
queryVector	number[]	Vector to search for.
topK	number	Number of results to return. Defaults to 10.
metadataFields	String[] or ALL_VECTOR_METADATA	A list of metadata fields to return with each result, or a value indicating all metadata should be returned. If not provided, no metadata is returned. Defaults to None.

Method response object

• Success
  • hits(): SearchAndFetchVectorsHit[]
    • SearchAndFetchVectorsHit:
      • id: string
      • score: number
      • metadata: Map<string, string>
      • vector: number[]
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.searchAndFetchVectors('test-index', [1.0, 2.0], {
  topK: 3,
  metadataFields: ALL_VECTOR_METADATA,
});
if (result instanceof VectorSearchAndFetchVectors.Success) {
  console.log(`Found ${result.hits().length} matches`);
} else if (result instanceof VectorSearchAndFetchVectors.Error) {
  throw new Error(`An error occurred searching index test-index: ${result.errorCode()}: ${result.toString()}`);
}

Python

response = await vector_client.search_and_fetch_vectors(
    "test-index", [1.0, 2.0], top_k=3, metadata_fields=ALL_METADATA
)
match response:
    case SearchAndFetchVectors.Success() as success:
        print(f"Found {len(success.hits)} matches")
    case SearchAndFetchVectors.Error() as error:
        print(f"Error searching index 'test-index': {error.message}")

Delete Item Batch

Deletes a batch of items from a vector index.

Name	Type	Description
indexName	String	Name of the vector index.
items	String[]	IDs of the items to be deleted.

Method response object

• Success
• Error

See response objects for specific information.

JavaScript

const result = await vectorClient.deleteItemBatch('test-index', ['example_item_1', 'example_item_2']);
if (result instanceof VectorUpsertItemBatch.Success) {
  console.log('Successfully deleted items');
} else if (result instanceof VectorUpsertItemBatch.Error) {
  throw new Error(`An error occurred while deleting items: ${result.errorCode()}: ${result.toString()}`);
}

Python

response = await vector_client.delete_item_batch("test-index", ["example_item_1", "example_item_2"])
match response:
    case DeleteItemBatch.Success():
        print("Successfully deleted items from index 'test-index'")
    case DeleteItemBatch.Error() as error:
        print(f"Error deleting items from index 'test-index': {error.message}")