Vectorize

Query and mutate a Cloudflare Vectorize index from Go.

Cloudflare Vectorize is a vector database for similarity search over embeddings. The exp/cloudflare/vectorize package wraps the bound index.

Declare the binding

[[vectorize]]
binding = "INDEX"
index_name = "my-index"

Resolve the binding

vectorize.NewVectorize resolves the binding by name and returns an error when no binding with that name exists.

import "github.com/syumai/workers-go/exp/cloudflare/vectorize"

index, err := vectorize.NewVectorize("INDEX")
if err != nil {
	// no binding named INDEX
}

Insert and upsert vectors

Insert and Upsert take a []VectorizeVector and return a VectorizeAsyncMutation — mutations are processed asynchronously and MutationID identifies the changeset. Insert errors when a provided ID already exists; Upsert replaces it.

mutation, err := index.Upsert([]vectorize.VectorizeVector{
	{ID: "1", Values: []float32{0.1, 0.2, 0.3}, Namespace: "docs"},
	{ID: "2", Values: []float32{0.4, 0.5, 0.6},
		Metadata: map[string]any{"title": "Getting started"}},
})
if err != nil {
	return err
}
fmt.Println(mutation.MutationID)

Query

Query runs a similarity search with a vector; QueryByID does the same starting from a vector already in the index. VectorizeQueryOptions controls TopK, Namespace, ReturnValues, ReturnMetadata (a js.Value), and Filter.

matches, err := index.Query(
	[]float32{0.1, 0.2, 0.3},
	vectorize.VectorizeQueryOptions{
		TopK:         5,
		ReturnValues: true,
	},
)
if err != nil {
	return err
}
for _, m := range matches.Matches {
	fmt.Println(m.ID, m.Score)
}

Fetch and delete by ID

vectors, err := index.GetByIds([]string{"1", "2"})

mutation, err := index.DeleteByIds([]string{"1"})

Describe returns a VectorizeIndexInfo with VectorCount, Dimensions, and the last processed mutation.