Vectorize Admin SDK

Use the Admin SDK when your backend needs to insert, update, search, or delete vectors in a declared Vectorize index.

Vectorize mutations are asynchronous. After insert, upsert, or delete, allow a few seconds before expecting queries to reflect the change. Metadata filtering beyond namespace also requires metadata indexes on the Cloudflare side.

Available Methods

Method	Purpose
upsert(vectors)	Insert new vectors or replace existing IDs
insert(vectors)	Insert only new vector IDs
search(vector, options?)	Search by raw vector values
queryById(vectorId, options?)	Search using an existing vector ID
getByIds(ids)	Fetch vectors by ID
delete(ids)	Remove vectors by ID
describe()	Read index metadata such as dimensions and metric

Language Method Matrix

The underlying operations are the same across SDKs. The main difference is naming style: camelCase, snake_case, Async, or Go-style capitalization.

Language	Methods	Notes
TypeScript / JavaScript	upsert, insert, search, queryById, getByIds, delete, describe	Same names in the JS admin SDK
Dart	upsert, insert, search, queryById, getByIds, delete, describe	Optional search params are named args
Kotlin	upsert, insert, search, queryById, getByIds, delete, describe	suspend methods
Java	upsert, insert, search, queryById, getByIds, delete, describe	Overloads for simple/full search options
Scala	upsert, insert, search, queryById, getByIds, delete, describe	Scala wrapper around Java admin SDK
Python	upsert, insert, search, query_by_id, get_by_ids, delete, describe	Snake_case for multi-word names
Go	Upsert, Insert, Search, QueryByID, GetByIDs, Delete, Describe	Context-first method signatures
PHP	upsert, insert, search, queryById, getByIds, delete, describe	PHP-style nullable optional args
Rust	upsert, insert, search, query_by_id, get_by_ids, delete, describe	Snake_case async methods
C#	UpsertAsync, InsertAsync, SearchAsync, QueryByIdAsync, GetByIdsAsync, DeleteAsync, DescribeAsync	CancellationToken optional
Ruby	upsert, insert, search, query_by_id, get_by_ids, delete, describe	Snake_case for multi-word names
Elixir	upsert, upsert!, insert, insert!, search, search!, query_by_id, query_by_id!, get_by_ids, get_by_ids!, delete, delete!, describe, describe!	! variants unwrap {:ok, ...}

TypeScript

import { createAdminClient } from '@edge-base/admin';

const admin = createAdminClient('https://your-app.example.com', {
  serviceKey: process.env.EDGEBASE_SERVICE_KEY!,
});

await admin.vector('embeddings').upsert([
  { id: 'doc-1', values: [0.1, 0.2], metadata: { title: 'Hello' } },
]);
const matches = await admin.vector('embeddings').search([0.1, 0.2], { topK: 10 });

Dart

import 'dart:io';
import 'package:edgebase_admin/edgebase_admin.dart';

final admin = AdminEdgeBase(
  'https://your-app.example.com',
  serviceKey: Platform.environment['EDGEBASE_SERVICE_KEY']!,
);

await admin.vector('embeddings').upsert([
  {'id': 'doc-1', 'values': [0.1, 0.2], 'metadata': {'title': 'Hello'}},
]);
final matches = await admin.vector('embeddings').search([0.1, 0.2], topK: 10);

Kotlin

import dev.edgebase.sdk.admin.AdminEdgeBase

val admin = AdminEdgeBase(
    "https://your-app.example.com",
    serviceKey = System.getenv("EDGEBASE_SERVICE_KEY") ?: ""
)

admin.vector("embeddings").upsert(listOf(
    mapOf("id" to "doc-1", "values" to listOf(0.1, 0.2), "metadata" to mapOf("title" to "Hello"))
))
val matches = admin.vector("embeddings").search(listOf(0.1, 0.2), topK = 10)

Java

import dev.edgebase.sdk.admin.*;
import java.util.List;
import java.util.Map;

AdminEdgeBase admin = EdgeBase.admin(
    "https://your-app.example.com",
    System.getenv("EDGEBASE_SERVICE_KEY")
);

admin.vector("embeddings").upsert(List.of(
    Map.of("id", "doc-1", "values", List.of(0.1, 0.2), "metadata", Map.of("title", "Hello"))
));
var matches = admin.vector("embeddings").search(List.of(0.1, 0.2), 10, null);

Scala

import dev.edgebase.sdk.scala.admin.AdminEdgeBase

val admin = AdminEdgeBase(
  "https://your-app.example.com",
  sys.env("EDGEBASE_SERVICE_KEY")
)

admin.vector("embeddings").upsert(Seq(
  Map("id" -> "doc-1", "values" -> Seq(0.1, 0.2), "metadata" -> Map("title" -> "Hello"))
))
val matches = admin.vector("embeddings").search(Seq(0.1, 0.2), topK = 10)

Python

import os
from edgebase_admin import create_admin_client

admin = create_admin_client(
    "https://your-app.example.com",
    service_key=os.environ["EDGEBASE_SERVICE_KEY"],
)

admin.vector("embeddings").upsert([
    {"id": "doc-1", "values": [0.1, 0.2], "metadata": {"title": "Hello"}},
])
matches = admin.vector("embeddings").search([0.1, 0.2], top_k=10)

Go

import (
    "context"
    "os"

    edgebase "github.com/edge-base/sdk-go"
)

ctx := context.Background()
admin := edgebase.NewAdminClient("https://your-app.example.com", os.Getenv("EDGEBASE_SERVICE_KEY"))

_, _ = admin.Vector("embeddings").Upsert(ctx, []map[string]interface{}{
    {"id": "doc-1", "values": []float64{0.1, 0.2}, "metadata": map[string]interface{}{"title": "Hello"}},
})
matches, _ := admin.Vector("embeddings").Search(ctx, []float64{0.1, 0.2}, &edgebase.VectorSearchOptions{TopK: 10})

PHP

use EdgeBase\Admin\AdminClient;

$admin = new AdminClient('https://your-app.example.com', getenv('EDGEBASE_SERVICE_KEY'));

$admin->vector('embeddings')->upsert([
    ['id' => 'doc-1', 'values' => [0.1, 0.2], 'metadata' => ['title' => 'Hello']],
]);
$matches = $admin->vector('embeddings')->search([0.1, 0.2], 10);

Rust

use edgebase_admin::EdgeBase;
use serde_json::json;

let admin = EdgeBase::server(
    "https://your-app.example.com",
    &std::env::var("EDGEBASE_SERVICE_KEY").unwrap(),
)?;

admin.vector("embeddings").upsert(&[json!({
    "id": "doc-1", "values": [0.1, 0.2], "metadata": {"title": "Hello"}
})]).await?;
let matches = admin.vector("embeddings").search(&[0.1, 0.2], 10, None, None, None, None).await?;

C#

using EdgeBase.Admin;

var admin = new AdminClient(
    "https://your-app.example.com",
    Environment.GetEnvironmentVariable("EDGEBASE_SERVICE_KEY")!
);

await admin.Vector("embeddings").UpsertAsync(new[] {
    new Dictionary<string, object?> { ["id"] = "doc-1", ["values"] = new[] { 0.1, 0.2 }, ["metadata"] = new Dictionary<string, object?> { ["title"] = "Hello" } },
});
var matches = await admin.Vector("embeddings").SearchAsync(new[] { 0.1, 0.2 }, topK: 10);

Ruby

require "edgebase_admin"

admin = EdgebaseAdmin::AdminClient.new(
  "https://your-app.example.com",
  service_key: ENV.fetch("EDGEBASE_SERVICE_KEY")
)

admin.vector("embeddings").upsert([
  { "id" => "doc-1", "values" => [0.1, 0.2], "metadata" => { "title" => "Hello" } }
])
matches = admin.vector("embeddings").search([0.1, 0.2], top_k: 10)

Elixir

alias EdgeBaseAdmin

admin =
  EdgeBaseAdmin.new("https://your-app.example.com",
    service_key: System.fetch_env!("EDGEBASE_SERVICE_KEY")
  )

vector = EdgeBaseAdmin.vector(admin, "embeddings")

EdgeBaseAdmin.Vector.upsert!(vector, [
  %{"id" => "doc-1", "values" => [0.1, 0.2], "metadata" => %{"title" => "Hello"}}
])
matches = EdgeBaseAdmin.Vector.search!(vector, [0.1, 0.2], top_k: 10)

Delete Vectors

Delete support is available through the Admin SDK and removes vectors by ID.

await admin.vector('embeddings').delete(['doc-1', 'doc-2']);

Like other Vectorize mutations, deletion is asynchronous and may take a few seconds to disappear from subsequent queries.