v2.0.0 Release

This version of the Pinecone .NET SDK introduces inference and imports. It also supports version 2024-10 of the Pinecone API. You can read more about versioning here.

Features

Embed

The Inference has an operation called Embed which allows users to generate embeddings for input data.

using Pinecone;

var pinecone = new PineconeClient("PINECONE_API_KEY");

// Prepare input sentences to be embedded
List<EmbedRequestInputsItem> inputs =
[
    new()
    {
        Text = "The quick brown fox jumps over the lazy dog."
    },
    new()
    {
        Text = "Lorem ipsum"
    }
];

// Specify the embedding model and parameters
var embeddingModel = "multilingual-e5-large";

// Generate embeddings for the input data
var embeddings = await pinecone.Inference.EmbedAsync(new EmbedRequest()
{
    Model = embeddingModel,
    Inputs = inputs,
    Parameters = new EmbedRequestParameters()
    {
        InputType = "query",
        Truncate = "END"
    }
});

// Get embedded data
var embeddedData = embeddings.Data;

Rerank

The Inference has another operation called Rerank which provides users the ability to rerank documents in descending relevance-order against a given query. Reranking documents is a common "second-pass" ranking strategy broadly used in retrieval applications.

using Pinecone;

var pinecone = new PineconeClient("PINECONE_API_KEY");

// The model to use for reranking
var model = "bge-reranker-v2-m3";

// The query to rerank documents against
var query = "The tech company Apple is known for its innovative products like the iPhone.";

// Add the documents to rerank
var documents = new List<Dictionary<string, string>>
{
    new()
    {
        ["id"] = "vec1",
        ["my_field"] = "Apple is a popular fruit known for its sweetness and crisp texture."
    },
    new()
    {
        ["id"] = "vec2",
        ["my_field"] = "Many people enjoy eating apples as a healthy snack."
    },
    new()
    {
        ["id"] = "vec3",
        ["my_field"] =
            "Apple Inc. has revolutionized the tech industry with its sleek designs and user-friendly interfaces."
    },
    new()
    {
        ["id"] = "vec4",
        ["my_field"] = "An apple a day keeps the doctor away, as the saying goes."
    }
};

// The fields to rank the documents by. If not provided, the default is "text"
var rankFields = new List<string> { "my_field" };

// The number of results to return sorted by relevance. Defaults to the number of inputs
int topN = 2;

// Whether to return the documents in the response
bool returnDocuments = true;

// Additional model-specific parameters for the reranker
var parameters = new Dictionary<string, string>
{
    ["truncate"] = "END"
};

// Send ranking request
var result = await pinecone.Inference.RerankAsync(
    new RerankRequest
    {
        Model = model,
        Query = query,
        Documents = documents,
        RankFields = rankFields,
        TopN = topN,
        Parameters = parameters
    });

// Get ranked data
var data = result.Data;

Import

Index now exposes additional methods for working with import operations. An import is a long-running, asynchronous operation that gives users the ability to import vectors directly from object storage (e.g. S3) into a Pinecone index. It is intended to be used with large-scale jobs. For small-scale jobs (e.g. <1000 vectors), we recommend continuing to use upsert.

using Pinecone;

...

// Initialize pinecone object
var pinecone = new PineconeClient("PINECONE_API_KEY");

// Get async imports connection object
var index = pinecone.Index("PINECONE_INDEX_NAME");

// s3 uri
var uri = "s3://path/to/file.parquet";

// Start an import
var startImportResponse = await index.StartBulkImportAsync(new StartImportRequest
{
    Uri = uri,
    IntegrationId = "123-456-789",
    ErrorMode = new ImportErrorMode { OnError = ImportErrorModeOnError.Continue }
});

// List imports
var listimportResponse = await index.ListBulkImportsAsync(new ListBulkImportsRequest
{
    Limit = 100,
    PaginationToken = "some-pagination-token"
});

// Describe import
var importDetails = await index.DescribeBulkImportAsync("1");

// Cancel import
var cancelResponse = await index.CancelBulkImportAsync("2");