Elasticsearch vector and full-text search in 10 minutes
Elasticsearch is a search and vector engine that supports full-text, semantic, and hybrid search over your data. This guide uses official Elasticsearch clients to take you from an empty Vector Database project to real search results in less than 10 minutes.
Use it if you're new to search in Elasticsearch or want to experiment with meaning-based retrieval alongside keyword matching.
Are you an agent? Use the elasticsearch-onboarding skill.
Create a free Elasticsearch Vector Database project. It's serverless and built for search and vector workloads, so you don't need to size or manage a cluster.
The project takes about a minute to start. The Getting started page then displays the Project endpoint and a generated API key. Copy both values.
You can also find the Project endpoint and API keys on the Home page. To find the Home page, use the global search field. Select Connection details to view your endpoint and manage API keys.
Set the project endpoint and API key as environment variables:
export ES_URL="<YOUR_PROJECT_URL>"
export ES_API_KEY="YOUR_API_KEY"
- For example,
https://my-project.es.us-east-1.aws.elastic.cloud:443.
$Env:ES_URL = "<YOUR_PROJECT_URL>"
$Env:ES_API_KEY = "YOUR_API_KEY"
- For example,
https://my-project.es.us-east-1.aws.elastic.cloud:443.
Install the client for your language. For a list of available clients, refer to Elasticsearch clients.
pip install elasticsearch
For supported Python versions and other requirements, refer to the Python client documentation.
npm install @elastic/elasticsearch
npm install --save-dev typescript tsx
npm pkg set type=module
For supported Node.js versions and other requirements, refer to the JavaScript client documentation.
composer require elasticsearch/elasticsearch
For supported PHP versions and other requirements, refer to the PHP client documentation.
bundle add elasticsearch
For supported Ruby versions and other requirements, refer to the Ruby client installation documentation.
dotnet add package Elastic.Clients.Elasticsearch
For supported .NET versions and other requirements, refer to the .NET client documentation.
Add the client dependency and the runnable example's main class to the generated build.gradle file:
dependencies {
implementation "co.elastic.clients:elasticsearch-java:VERSION"
}
application {
mainClass = "quickstart.App"
}
Replace VERSION with the version from the latest Java client release. For supported Java versions and other requirements, refer to the Java client documentation.
go get github.com/elastic/go-elasticsearch/v9
For supported Go versions and other requirements, refer to the Go client installation documentation.
After installation, initialize the client:
import os
from elasticsearch import Elasticsearch, helpers
es = Elasticsearch(
os.environ["ES_URL"],
api_key=os.environ["ES_API_KEY"],
)
import { Client } from "@elastic/elasticsearch";
const es = new Client({
node: process.env.ES_URL,
auth: { apiKey: process.env.ES_API_KEY! },
});
<?php
require __DIR__ . "/vendor/autoload.php";
use Elastic\Elasticsearch\ClientBuilder;
$es = ClientBuilder::create()
->setHosts([getenv("ES_URL")])
->setApiKey(getenv("ES_API_KEY"))
->build();
require "elasticsearch"
es = Elasticsearch::Client.new(
url: ENV.fetch("ES_URL"),
api_key: ENV.fetch("ES_API_KEY")
)
using System.Text.Json;
using System.Text.Json.Serialization;
using Elastic.Clients.Elasticsearch;
using Elastic.Clients.Elasticsearch.Core.Bulk;
using Elastic.Clients.Elasticsearch.Esql;
using Elastic.Transport;
var url = Environment.GetEnvironmentVariable("ES_URL")
?? throw new InvalidOperationException("ES_URL is not set.");
var apiKey = Environment.GetEnvironmentVariable("ES_API_KEY")
?? throw new InvalidOperationException("ES_API_KEY is not set.");
var settings = new ElasticsearchClientSettings(new Uri(url))
.Authentication(new ApiKey(apiKey));
var es = new ElasticsearchClient(settings);
package quickstart;
import java.io.BufferedReader;
import java.io.InputStreamReader;
import java.io.StringReader;
import co.elastic.clients.elasticsearch.ElasticsearchClient;
import co.elastic.clients.elasticsearch._types.Refresh;
import co.elastic.clients.elasticsearch.core.BulkRequest;
import co.elastic.clients.elasticsearch.core.SearchResponse;
import co.elastic.clients.elasticsearch.core.search.Hit;
import co.elastic.clients.elasticsearch.esql.EsqlFormat;
import co.elastic.clients.transport.endpoints.BinaryResponse;
public class App {
public static void main(String[] args) throws Exception {
String serverUrl = System.getenv("ES_URL");
String apiKey = System.getenv("ES_API_KEY");
ElasticsearchClient es = ElasticsearchClient.of(b -> b
.host(serverUrl)
.apiKey(apiKey)
);
}
}
package main
import (
"log"
"os"
"github.com/elastic/go-elasticsearch/v9"
)
func main() {
es, err := elasticsearch.New(
elasticsearch.WithAddresses(os.Getenv("ES_URL")),
elasticsearch.WithAPIKey(os.Getenv("ES_API_KEY")),
)
if err != nil {
log.Fatal(err)
}
}
Create the books index and add the sample data.
es.indices.create(
index="books",
mappings={
"properties": {
"description": {
"type": "semantic_text",
}
}
},
)
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. semantic_textautomatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
In this example,semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
await es.indices.create({
index: "books",
mappings: {
properties: {
description: {
type: "semantic_text",
},
},
},
});
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
$es->indices()->create([
"index" => "books",
"body" => [
"mappings" => [
"properties" => [
"description" => [
"type" => "semantic_text",
],
],
],
],
]);
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
es.indices.create(
index: "books",
body: {
mappings: {
properties: {
description: {
type: "semantic_text"
}
}
}
}
)
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
await es.Indices.CreateAsync<Book>("books", c => c
.Mappings(m => m
.Properties(p => p
.SemanticText(b => b.Description)
)
)
);
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
es.indices().create(c -> c
.index("books")
.withJson(new StringReader("""
{
"mappings": {
"properties": {
"description": {
"type": "semantic_text"
}
}
}
}
"""))
);
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
createResponse, err := es.Indices.Create(
"books",
es.Indices.Create.WithBody(strings.NewReader(`{
"mappings": {
"properties": {
"description": {
"type": "semantic_text"
}
}
}
}`)),
)
if err != nil {
panic(err)
}
if createResponse.IsError() {
log.Fatal(createResponse)
}
createResponse.Body.Close()
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
PUT books
{
"mappings": {
"properties": {
"description": {
"type": "semantic_text"
}
}
}
}
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_textautomatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
curl -X PUT "$ES_URL/books" \
-H "Authorization: ApiKey $ES_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"mappings": {
"properties": {
"description": {
"type": "semantic_text"
}
}
}
}'
- Elasticsearch uses dynamic mapping to determine field types from the first documents you index. The only field you need to define is
description, which you set tosemantic_textbefore indexing so you can search it by meaning. - In this example,
semantic_textuses a default inference endpoint. The model used by this endpoint is multilingual, so you can index and search text in multiple languages. You can also configure inference endpoints forsemantic_textto use custom models.
semantic_text automatically embeds text at ingest time using an inference endpoint. The inference endpoint connects to an embedding model that converts indexed text and search queries into vectors.
Next, index five books in one bulk request:
books = [
{"title": "The Left Hand of Darkness", "author": "Ursula K. Le Guin", "release_year": 1969,
"description": "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."},
{"title": "Project Hail Mary", "author": "Andy Weir", "release_year": 2021,
"description": "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."},
{"title": "The Name of the Wind", "author": "Patrick Rothfuss", "release_year": 2007,
"description": "A gifted young musician and magician tells the story of his rise from orphan to legend."},
{"title": "Klara and the Sun", "author": "Kazuo Ishiguro", "release_year": 2021,
"description": "An artificial friend watches human love and loneliness while hoping a child will pick her."},
{"title": "Dune", "author": "Frank Herbert", "release_year": 1965,
"description": "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."},
]
helpers.bulk(es, ({"_index": "books", "_source": b} for b in books), refresh="wait_for")
const books = [
{
title: "The Left Hand of Darkness",
author: "Ursula K. Le Guin",
release_year: 1969,
description: "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap.",
},
{
title: "Project Hail Mary",
author: "Andy Weir",
release_year: 2021,
description: "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth.",
},
{
title: "The Name of the Wind",
author: "Patrick Rothfuss",
release_year: 2007,
description: "A gifted young musician and magician tells the story of his rise from orphan to legend.",
},
{
title: "Klara and the Sun",
author: "Kazuo Ishiguro",
release_year: 2021,
description: "An artificial friend watches human love and loneliness while hoping a child will pick her.",
},
{
title: "Dune",
author: "Frank Herbert",
release_year: 1965,
description: "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power.",
},
];
await es.bulk({
refresh: "wait_for",
operations: books.flatMap((book) => [
{ index: { _index: "books" } },
book,
]),
});
$books = [
[
"title" => "The Left Hand of Darkness",
"author" => "Ursula K. Le Guin",
"release_year" => 1969,
"description" => "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap.",
],
[
"title" => "Project Hail Mary",
"author" => "Andy Weir",
"release_year" => 2021,
"description" => "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth.",
],
[
"title" => "The Name of the Wind",
"author" => "Patrick Rothfuss",
"release_year" => 2007,
"description" => "A gifted young musician and magician tells the story of his rise from orphan to legend.",
],
[
"title" => "Klara and the Sun",
"author" => "Kazuo Ishiguro",
"release_year" => 2021,
"description" => "An artificial friend watches human love and loneliness while hoping a child will pick her.",
],
[
"title" => "Dune",
"author" => "Frank Herbert",
"release_year" => 1965,
"description" => "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power.",
],
];
$operations = [];
foreach ($books as $book) {
$operations[] = [
"index" => [
"_index" => "books",
],
];
$operations[] = $book;
}
$es->bulk([
"refresh" => "wait_for",
"body" => $operations,
]);
books = [
{
title: "The Left Hand of Darkness",
author: "Ursula K. Le Guin",
release_year: 1969,
description: "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."
},
{
title: "Project Hail Mary",
author: "Andy Weir",
release_year: 2021,
description: "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."
},
{
title: "The Name of the Wind",
author: "Patrick Rothfuss",
release_year: 2007,
description: "A gifted young musician and magician tells the story of his rise from orphan to legend."
},
{
title: "Klara and the Sun",
author: "Kazuo Ishiguro",
release_year: 2021,
description: "An artificial friend watches human love and loneliness while hoping a child will pick her."
},
{
title: "Dune",
author: "Frank Herbert",
release_year: 1965,
description: "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."
}
]
operations = books.flat_map do |book|
[
{ index: { _index: "books" } },
book
]
end
es.bulk(
body: operations,
refresh: "wait_for"
)
var books = new[]
{
new Book(
"The Left Hand of Darkness",
"Ursula K. Le Guin",
1969,
"An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."
),
new Book(
"Project Hail Mary",
"Andy Weir",
2021,
"A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."
),
new Book(
"The Name of the Wind",
"Patrick Rothfuss",
2007,
"A gifted young musician and magician tells the story of his rise from orphan to legend."
),
new Book(
"Klara and the Sun",
"Kazuo Ishiguro",
2021,
"An artificial friend watches human love and loneliness while hoping a child will pick her."
),
new Book(
"Dune",
"Frank Herbert",
1965,
"On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."
)
};
var operations = new BulkOperationsCollection();
foreach (var book in books)
{
operations.Add(new BulkIndexOperation<Book>(book)
{
Index = "books"
});
}
await es.BulkAsync(new BulkRequest
{
Refresh = Refresh.WaitFor,
Operations = operations
});
public record Book(
string Title,
string Author,
[property: JsonPropertyName("release_year")] int ReleaseYear,
string Description
);
record Book(
String title,
String author,
int release_year,
String description
) {}
Book[] books = {
new Book(
"The Left Hand of Darkness",
"Ursula K. Le Guin",
1969,
"An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."
),
new Book(
"Project Hail Mary",
"Andy Weir",
2021,
"A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."
),
new Book(
"The Name of the Wind",
"Patrick Rothfuss",
2007,
"A gifted young musician and magician tells the story of his rise from orphan to legend."
),
new Book(
"Klara and the Sun",
"Kazuo Ishiguro",
2021,
"An artificial friend watches human love and loneliness while hoping a child will pick her."
),
new Book(
"Dune",
"Frank Herbert",
1965,
"On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."
)
};
BulkRequest.Builder bulk = new BulkRequest.Builder()
.refresh(Refresh.WaitFor);
for (Book book : books) {
bulk.operations(op -> op
.index(idx -> idx
.index("books")
.document(book)
)
);
}
es.bulk(bulk.build());
type Book struct {
Title string `json:"title"`
Author string `json:"author"`
ReleaseYear int `json:"release_year"`
Description string `json:"description"`
}
books := []Book{
{
Title: "The Left Hand of Darkness",
Author: "Ursula K. Le Guin",
ReleaseYear: 1969,
Description: "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap.",
},
{
Title: "Project Hail Mary",
Author: "Andy Weir",
ReleaseYear: 2021,
Description: "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth.",
},
{
Title: "The Name of the Wind",
Author: "Patrick Rothfuss",
ReleaseYear: 2007,
Description: "A gifted young musician and magician tells the story of his rise from orphan to legend.",
},
{
Title: "Klara and the Sun",
Author: "Kazuo Ishiguro",
ReleaseYear: 2021,
Description: "An artificial friend watches human love and loneliness while hoping a child will pick her.",
},
{
Title: "Dune",
Author: "Frank Herbert",
ReleaseYear: 1965,
Description: "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power.",
},
}
indexer, err := esutil.NewBulkIndexer(esutil.BulkIndexerConfig{
Client: es,
Index: "books",
Refresh: "wait_for",
})
if err != nil {
panic(err)
}
ctx := context.Background()
for _, book := range books {
data, err := json.Marshal(book)
if err != nil {
panic(err)
}
err = indexer.Add(ctx, esutil.BulkIndexerItem{
Action: "index",
Body: bytes.NewReader(data),
})
if err != nil {
panic(err)
}
}
if err := indexer.Close(ctx); err != nil {
panic(err)
}
POST books/_bulk?refresh=wait_for
{ "index": {} }
{ "title": "The Left Hand of Darkness", "author": "Ursula K. Le Guin", "release_year": 1969, "description": "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap." }
{ "index": {} }
{ "title": "Project Hail Mary", "author": "Andy Weir", "release_year": 2021, "description": "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth." }
{ "index": {} }
{ "title": "The Name of the Wind", "author": "Patrick Rothfuss", "release_year": 2007, "description": "A gifted young musician and magician tells the story of his rise from orphan to legend." }
{ "index": {} }
{ "title": "Klara and the Sun", "author": "Kazuo Ishiguro", "release_year": 2021, "description": "An artificial friend watches human love and loneliness while hoping a child will pick her." }
{ "index": {} }
{ "title": "Dune", "author": "Frank Herbert", "release_year": 1965, "description": "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power." }
curl -X POST "$ES_URL/books/_bulk?refresh=wait_for" \
-H "Authorization: ApiKey $ES_API_KEY" \
-H "Content-Type: application/x-ndjson" \
--data-binary @- <<'EOF'
{ "index": {} }
{ "title": "The Left Hand of Darkness", "author": "Ursula K. Le Guin", "release_year": 1969, "description": "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap." }
{ "index": {} }
{ "title": "Project Hail Mary", "author": "Andy Weir", "release_year": 2021, "description": "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth." }
{ "index": {} }
{ "title": "The Name of the Wind", "author": "Patrick Rothfuss", "release_year": 2007, "description": "A gifted young musician and magician tells the story of his rise from orphan to legend." }
{ "index": {} }
{ "title": "Klara and the Sun", "author": "Kazuo Ishiguro", "release_year": 2021, "description": "An artificial friend watches human love and loneliness while hoping a child will pick her." }
{ "index": {} }
{ "title": "Dune", "author": "Frank Herbert", "release_year": 1965, "description": "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power." }
EOF
Use the indexed book data to run semantic and hybrid searches. Semantic search matches meaning, while hybrid search combines semantic and keyword matching.
Run a semantic search against the description field:
resp = es.search(
index="books",
query={
"semantic": {
"field": "description",
"query": "surviving alone in space",
}
},
)
for hit in resp["hits"]["hits"]:
print(hit["_score"], hit["_source"]["title"])
interface Book {
title: string;
}
const resp = await es.search<Book>({
index: "books",
query: {
semantic: {
field: "description",
query: "surviving alone in space",
},
},
});
for (const hit of resp.hits.hits) {
console.log(hit._score, hit._source?.title);
}
$response = $es->search([
"index" => "books",
"body" => [
"query" => [
"semantic" => [
"field" => "description",
"query" => "surviving alone in space",
],
],
],
]);
foreach ($response["hits"]["hits"] as $hit) {
echo $hit["_score"] . " " . $hit["_source"]["title"] . "\n";
}
response = es.search(
index: "books",
body: {
query: {
semantic: {
field: "description",
query: "surviving alone in space"
}
}
}
)
response["hits"]["hits"].each do |hit|
puts "#{hit["_score"]} #{hit["_source"]["title"]}"
end
var response = await es.SearchAsync<SearchBook>(s => s
.Indices("books")
.Query(q => q
.Semantic(semantic => semantic
.Field("description")
.Query("surviving alone in space")
)
)
);
foreach (var hit in response.Hits)
{
Console.WriteLine($"{hit.Score} {hit.Source?.Title}");
}
public record SearchBook(
[property: JsonPropertyName("title")] string Title
);
String query = """
{
"query": {
"semantic": {
"field": "description",
"query": "surviving alone in space"
}
}
}
""";
SearchResponse<Book> response = es.search(s -> s
.index("books")
.withJson(new StringReader(query)),
Book.class
);
for (Hit<Book> hit : response.hits().hits()) {
System.out.println(hit.score() + " " + hit.source().title());
}
query := strings.NewReader(`{
"query": {
"semantic": {
"field": "description",
"query": "surviving alone in space"
}
}
}`)
res, err := es.Search(
es.Search.WithContext(context.Background()),
es.Search.WithIndex("books"),
es.Search.WithBody(query),
)
if err != nil {
log.Fatal(err)
}
defer res.Body.Close()
if res.IsError() {
log.Fatal(res)
}
var response struct {
Hits struct {
Hits []struct {
Score float64 `json:"_score"`
Source struct {
Title string `json:"title"`
} `json:"_source"`
} `json:"hits"`
} `json:"hits"`
}
if err := json.NewDecoder(res.Body).Decode(&response); err != nil {
log.Fatal(err)
}
for _, hit := range response.Hits.Hits {
fmt.Println(hit.Score, hit.Source.Title)
}
GET books/_search
{
"query": {
"semantic": {
"field": "description",
"query": "surviving alone in space"
}
}
}
curl -X GET "$ES_URL/books/_search" \
-H "Authorization: ApiKey $ES_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": {
"semantic": {
"field": "description",
"query": "surviving alone in space"
}
}
}'
Example output
The first result is:
SCORE Project Hail Mary
Project Hail Mary ranks first because its description includes "A lone astronaut," which is semantically similar to "surviving alone in space" even though the wording differs.
Run a hybrid search that uses full-text search on the title field and semantic search on the description field:
resp = es.search(
index="books",
query={
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age",
}
},
]
}
},
)
for hit in resp["hits"]["hits"]:
print(hit["_score"], hit["_source"]["title"])
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
const resp = await es.search<Book>({
index: "books",
query: {
bool: {
should: [
{
match: {
title: "wind",
},
},
{
semantic: {
field: "description",
query: "young magician coming of age",
},
},
],
},
},
});
for (const hit of resp.hits.hits) {
console.log(hit._score, hit._source?.title);
}
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
$response = $es->search([
"index" => "books",
"body" => [
"query" => [
"bool" => [
"should" => [
[
"match" => [
"title" => "wind",
],
],
[
"semantic" => [
"field" => "description",
"query" => "young magician coming of age",
],
],
],
],
],
],
]);
foreach ($response["hits"]["hits"] as $hit) {
echo $hit["_score"] . " " . $hit["_source"]["title"] . "\n";
}
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
response = es.search(
index: "books",
body: {
query: {
bool: {
should: [
{
match: {
title: "wind"
}
},
{
semantic: {
field: "description",
query: "young magician coming of age"
}
}
]
}
}
}
)
response["hits"]["hits"].each do |hit|
puts "#{hit["_score"]} #{hit["_source"]["title"]}"
end
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
var response = await es.SearchAsync<SearchBook>(s => s
.Indices("books")
.Query(q => q
.Bool(b => b
.Should(
should => should
.Match(match => match
.Field(book => book.Title)
.Query("wind")
),
should => should
.Semantic(semantic => semantic
.Field("description")
.Query("young magician coming of age")
)
)
)
)
);
foreach (var hit in response.Hits)
{
Console.WriteLine($"{hit.Score} {hit.Source?.Title}");
}
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
String query = """
{
"query": {
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age"
}
}
]
}
}
}
""";
SearchResponse<Book> response = es.search(s -> s
.index("books")
.withJson(new StringReader(query)),
Book.class
);
for (Hit<Book> hit : response.hits().hits()) {
System.out.println(hit.score() + " " + hit.source().title());
}
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
query := strings.NewReader(`{
"query": {
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age"
}
}
]
}
}
}`)
res, err := es.Search(
es.Search.WithContext(context.Background()),
es.Search.WithIndex("books"),
es.Search.WithBody(query),
)
if err != nil {
log.Fatal(err)
}
defer res.Body.Close()
if res.IsError() {
log.Fatal(res)
}
var response struct {
Hits struct {
Hits []struct {
Score float64 `json:"_score"`
Source struct {
Title string `json:"title"`
} `json:"_source"`
} `json:"hits"`
} `json:"hits"`
}
if err := json.NewDecoder(res.Body).Decode(&response); err != nil {
log.Fatal(err)
}
for _, hit := range response.Hits.Hits {
fmt.Println(hit.Score, hit.Source.Title)
}
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
GET books/_search
{
"query": {
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age"
}
}
]
}
}
}
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
curl -X GET "$ES_URL/books/_search" \
-H "Authorization: ApiKey $ES_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": {
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age"
}
}
]
}
}
}'
- The
matchclause performs full-text search ontitleand scores how well the analyzed text matcheswind. - The
semanticclause searchesdescriptionby meaning and contributes its semantic similarity score.
Example output
The first result is:
SCORE The Name of the Wind
The Name of the Wind ranks first because its description of a gifted young magician's rise from orphan to legend is semantically similar to young magician coming of age.
With aggregations, you can summarize groups of data, such as the number of books released in each decade. Use ES|QL, a piped query language, to run this aggregation:
resp = es.esql.query(query="""
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| SORT decade ASC
| LIMIT 10
""")
cols = {c["name"]: i for i, c in enumerate(resp["columns"])}
for row in resp["values"]:
print(f"{row[cols['decade']]}s: {row[cols['books']]}")
const resp = await es.esql.query({
query: `
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
`,
});
for (const [decade, count] of resp.values ?? []) {
console.log(`${decade}s: ${count}`);
}
$response = $es->esql()->query([
"body" => [
"query" => <<<'ESQL'
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
ESQL,
],
]);
foreach ($response["values"] as [$decade, $count]) {
echo $decade . "s: " . $count . "\n";
}
response = es.esql.query(
body: {
query: <<~ESQL
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
ESQL
}
)
response["values"].each do |decade, count|
puts "#{decade}s: #{count}"
end
var response = await es.Esql.QueryAsync(r => r
.Query("""
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
""")
.Format(EsqlFormat.Json)
);
using var result = JsonDocument.Parse(response.Body);
foreach (
var row in result.RootElement.GetProperty("values").EnumerateArray()
)
{
Console.WriteLine($"{row[0]}s: {row[1]}");
}
String query = """
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
""";
BinaryResponse response = es.esql().query(q -> q
.query(query)
.format(EsqlFormat.Csv)
);
try (BufferedReader reader = new BufferedReader(
new InputStreamReader(response.content())
)) {
reader.lines().skip(1).forEach(line -> {
String[] values = line.split(",", -1);
System.out.println(values[0] + "s: " + values[1]);
});
}
typed, err := elasticsearch.NewTyped(
elasticsearch.WithAddresses(os.Getenv("ES_URL")),
elasticsearch.WithAPIKey(os.Getenv("ES_API_KEY")),
)
if err != nil {
log.Fatal(err)
}
query := `FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC`
response, err := typed.Esql.Query().
Query(query).
Format(esqlformat.Csv).
Do(context.Background())
if err != nil {
log.Fatal(err)
}
rows, err := csv.NewReader(bytes.NewReader(response)).ReadAll()
if err != nil {
log.Fatal(err)
}
for _, row := range rows[1:] {
fmt.Printf("%ss: %s\n", row[0], row[1])
}
POST _query
{
"query": """
FROM books
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
"""
}
curl -X POST "$ES_URL/_query" \
-H "Authorization: ApiKey $ES_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "FROM books | STATS books = COUNT(*) BY decade = release_year - (release_year % 10) | KEEP decade, books | SORT decade ASC"
}'
Example output
The output contains one row per decade:
1960s: 2
2000s: 1
2020s: 2
- Create a Vector Database project.
- Connect to your project.
- Install and initialize the client for your language.
- Select your language, expand the complete script, and run it.
Complete runnable script
import os
from elasticsearch import Elasticsearch, helpers
es = Elasticsearch(os.environ["ES_URL"], api_key=os.environ["ES_API_KEY"])
# Declare the semantically searchable field. Other fields are inferred on ingest.
es.indices.create(
index="books-complete",
mappings={"properties": {"description": {"type": "semantic_text"}}},
)
# Index the sample data.
books = [
{"title": "The Left Hand of Darkness", "author": "Ursula K. Le Guin", "release_year": 1969,
"description": "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."},
{"title": "Project Hail Mary", "author": "Andy Weir", "release_year": 2021,
"description": "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."},
{"title": "The Name of the Wind", "author": "Patrick Rothfuss", "release_year": 2007,
"description": "A gifted young musician and magician tells the story of his rise from orphan to legend."},
{"title": "Klara and the Sun", "author": "Kazuo Ishiguro", "release_year": 2021,
"description": "An artificial friend watches human love and loneliness while hoping a child will pick her."},
{"title": "Dune", "author": "Frank Herbert", "release_year": 1965,
"description": "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."},
]
helpers.bulk(es, ({"_index": "books-complete", "_source": b} for b in books), refresh="wait_for")
# Semantic search.
resp = es.search(
index="books-complete",
query={"semantic": {"field": "description", "query": "surviving alone in space"}},
)
print("=== Semantic search results ===")
for hit in resp["hits"]["hits"]:
print(" ", hit["_score"], hit["_source"]["title"])
# Hybrid search with keyword and semantic queries.
resp = es.search(
index="books-complete",
query={"bool": {"should": [
{"match": {"title": "wind"}},
{"semantic": {"field": "description", "query": "young magician coming of age"}},
]}},
)
print("\n=== Hybrid search results ===")
for hit in resp["hits"]["hits"]:
print(" ", hit["_score"], hit["_source"]["title"])
# Count books per decade.
resp = es.esql.query(query="""
FROM "books-complete"
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| SORT decade ASC
| LIMIT 10
""")
cols = {c["name"]: i for i, c in enumerate(resp["columns"])}
print("\n=== Books by decade ===")
for row in resp["values"]:
print(f"{row[cols['decade']]}s: {row[cols['books']]}")
Complete runnable script
import { Client } from "@elastic/elasticsearch";
interface Book {
title: string;
}
const es = new Client({
node: process.env.ES_URL,
auth: { apiKey: process.env.ES_API_KEY! },
});
// Declare the semantically searchable field. Other fields are inferred on ingest.
await es.indices.create({
index: "books-complete",
mappings: { properties: { description: { type: "semantic_text" } } },
});
// Index the sample data.
const books = [
{ title: "The Left Hand of Darkness", author: "Ursula K. Le Guin", release_year: 1969,
description: "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap." },
{ title: "Project Hail Mary", author: "Andy Weir", release_year: 2021,
description: "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth." },
{ title: "The Name of the Wind", author: "Patrick Rothfuss", release_year: 2007,
description: "A gifted young musician and magician tells the story of his rise from orphan to legend." },
{ title: "Klara and the Sun", author: "Kazuo Ishiguro", release_year: 2021,
description: "An artificial friend watches human love and loneliness while hoping a child will pick her." },
{ title: "Dune", author: "Frank Herbert", release_year: 1965,
description: "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power." },
];
await es.helpers.bulk({
datasource: books,
onDocument: () => ({ index: { _index: "books-complete" } }),
refreshOnCompletion: true,
});
// Semantic search.
let resp = await es.search<Book>({
index: "books-complete",
query: { semantic: { field: "description", query: "surviving alone in space" } },
});
console.log("=== Semantic search results ===");
for (const hit of resp.hits.hits) {
console.log(" ", hit._score, hit._source?.title);
}
// Hybrid search with keyword and semantic queries.
resp = await es.search<Book>({
index: "books-complete",
query: { bool: { should: [
{ match: { title: "wind" } },
{ semantic: { field: "description", query: "young magician coming of age" } },
] } },
});
console.log("\n=== Hybrid search results ===");
for (const hit of resp.hits.hits) {
console.log(" ", hit._score, hit._source?.title);
}
// Count books per decade.
const agg = await es.esql.query({ query: `
FROM "books-complete"
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
` });
console.log("\n=== Books by decade ===", agg.values);
Complete runnable script
<?php
require __DIR__ . "/vendor/autoload.php";
use Elastic\Elasticsearch\ClientBuilder;
$es = ClientBuilder::create()
->setHosts([getenv("ES_URL")])
->setApiKey(getenv("ES_API_KEY"))
->build();
$es->indices()->create([
"index" => "books-complete",
"body" => [
"mappings" => [
"properties" => [
"description" => [
"type" => "semantic_text",
],
],
],
],
]);
$books = [
[
"title" => "The Left Hand of Darkness",
"author" => "Ursula K. Le Guin",
"release_year" => 1969,
"description" => "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap.",
],
[
"title" => "Project Hail Mary",
"author" => "Andy Weir",
"release_year" => 2021,
"description" => "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth.",
],
[
"title" => "The Name of the Wind",
"author" => "Patrick Rothfuss",
"release_year" => 2007,
"description" => "A gifted young musician and magician tells the story of his rise from orphan to legend.",
],
[
"title" => "Klara and the Sun",
"author" => "Kazuo Ishiguro",
"release_year" => 2021,
"description" => "An artificial friend watches human love and loneliness while hoping a child will pick her.",
],
[
"title" => "Dune",
"author" => "Frank Herbert",
"release_year" => 1965,
"description" => "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power.",
],
];
$operations = [];
foreach ($books as $book) {
$operations[] = ["index" => ["_index" => "books-complete"]];
$operations[] = $book;
}
$es->bulk([
"refresh" => "wait_for",
"body" => $operations,
]);
$response = $es->search([
"index" => "books-complete",
"body" => [
"query" => [
"semantic" => [
"field" => "description",
"query" => "surviving alone in space",
],
],
],
]);
echo "=== Semantic search results ===\n";
foreach ($response["hits"]["hits"] as $hit) {
echo " " . $hit["_score"] . " " . $hit["_source"]["title"] . "\n";
}
$response = $es->search([
"index" => "books-complete",
"body" => [
"query" => [
"bool" => [
"should" => [
["match" => ["title" => "wind"]],
["semantic" => [
"field" => "description",
"query" => "young magician coming of age",
]],
],
],
],
],
]);
echo "\n=== Hybrid search results ===\n";
foreach ($response["hits"]["hits"] as $hit) {
echo " " . $hit["_score"] . " " . $hit["_source"]["title"] . "\n";
}
$response = $es->esql()->query([
"body" => [
"query" => <<<'ESQL'
FROM "books-complete"
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
ESQL,
],
]);
echo "\n=== Books by decade ===\n";
foreach ($response["values"] as [$decade, $count]) {
echo " " . $decade . "s: " . $count . "\n";
}
Complete runnable script
require "elasticsearch"
es = Elasticsearch::Client.new(
url: ENV.fetch("ES_URL"),
api_key: ENV.fetch("ES_API_KEY")
)
es.indices.create(
index: "books-complete",
body: {
mappings: {
properties: {
description: {
type: "semantic_text"
}
}
}
}
)
books = [
{
title: "The Left Hand of Darkness",
author: "Ursula K. Le Guin",
release_year: 1969,
description: "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."
},
{
title: "Project Hail Mary",
author: "Andy Weir",
release_year: 2021,
description: "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."
},
{
title: "The Name of the Wind",
author: "Patrick Rothfuss",
release_year: 2007,
description: "A gifted young musician and magician tells the story of his rise from orphan to legend."
},
{
title: "Klara and the Sun",
author: "Kazuo Ishiguro",
release_year: 2021,
description: "An artificial friend watches human love and loneliness while hoping a child will pick her."
},
{
title: "Dune",
author: "Frank Herbert",
release_year: 1965,
description: "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."
}
]
operations = books.flat_map do |book|
[
{ index: { _index: "books-complete" } },
book
]
end
es.bulk(body: operations, refresh: "wait_for")
response = es.search(
index: "books-complete",
body: {
query: {
semantic: {
field: "description",
query: "surviving alone in space"
}
}
}
)
puts "=== Semantic search results ==="
response["hits"]["hits"].each do |hit|
puts " #{hit["_score"]} #{hit["_source"]["title"]}"
end
response = es.search(
index: "books-complete",
body: {
query: {
bool: {
should: [
{ match: { title: "wind" } },
{
semantic: {
field: "description",
query: "young magician coming of age"
}
}
]
}
}
}
)
puts "\n=== Hybrid search results ==="
response["hits"]["hits"].each do |hit|
puts " #{hit["_score"]} #{hit["_source"]["title"]}"
end
response = es.esql.query(
body: {
query: <<~ESQL
FROM "books-complete"
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
ESQL
}
)
puts "\n=== Books by decade ==="
response["values"].each do |decade, count|
puts " #{decade}s: #{count}"
end
Complete runnable script
using System.Text.Json;
using System.Text.Json.Serialization;
using Elastic.Clients.Elasticsearch;
using Elastic.Clients.Elasticsearch.Core.Bulk;
using Elastic.Clients.Elasticsearch.Esql;
using Elastic.Transport;
var url = Environment.GetEnvironmentVariable("ES_URL")
?? throw new InvalidOperationException("ES_URL is not set.");
var apiKey = Environment.GetEnvironmentVariable("ES_API_KEY")
?? throw new InvalidOperationException("ES_API_KEY is not set.");
var settings = new ElasticsearchClientSettings(new Uri(url))
.Authentication(new ApiKey(apiKey));
var es = new ElasticsearchClient(settings);
await es.Indices.CreateAsync<Book>("books-complete", c => c
.Mappings(m => m
.Properties(p => p
.SemanticText(b => b.Description)
)
)
);
var books = new List<Book>
{
new(
"The Left Hand of Darkness",
"Ursula K. Le Guin",
1969,
"An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."
),
new(
"Project Hail Mary",
"Andy Weir",
2021,
"A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."
),
new(
"The Name of the Wind",
"Patrick Rothfuss",
2007,
"A gifted young musician and magician tells the story of his rise from orphan to legend."
),
new(
"Klara and the Sun",
"Kazuo Ishiguro",
2021,
"An artificial friend watches human love and loneliness while hoping a child will pick her."
),
new(
"Dune",
"Frank Herbert",
1965,
"On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."
)
};
var operations = new BulkOperationsCollection();
foreach (var book in books)
{
operations.Add(new BulkIndexOperation<Book>(book)
{
Index = "books-complete"
});
}
await es.BulkAsync(new BulkRequest
{
Refresh = Refresh.WaitFor,
Operations = operations
});
var response = await es.SearchAsync<Book>(s => s
.Indices("books-complete")
.Query(q => q
.Semantic(semantic => semantic
.Field("description")
.Query("surviving alone in space")
)
)
);
Console.WriteLine("=== Semantic search results ===");
foreach (var hit in response.Hits)
{
Console.WriteLine($" {hit.Score} {hit.Source?.Title}");
}
response = await es.SearchAsync<Book>(s => s
.Indices("books-complete")
.Query(q => q
.Bool(b => b
.Should(
should => should
.Match(match => match
.Field(book => book.Title)
.Query("wind")
),
should => should
.Semantic(semantic => semantic
.Field("description")
.Query("young magician coming of age")
)
)
)
)
);
Console.WriteLine("\n=== Hybrid search results ===");
foreach (var hit in response.Hits)
{
Console.WriteLine($" {hit.Score} {hit.Source?.Title}");
}
var aggregation = await es.Esql.QueryAsync(r => r
.Query("""
FROM "books-complete"
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
""")
.Format(EsqlFormat.Json)
);
using var result = JsonDocument.Parse(aggregation.Body);
Console.WriteLine("\n=== Books by decade ===");
foreach (var row in result.RootElement.GetProperty("values").EnumerateArray())
{
Console.WriteLine($" {row[0]}s: {row[1]}");
}
public record Book(
[property: JsonPropertyName("title")] string Title,
[property: JsonPropertyName("author")] string Author,
[property: JsonPropertyName("release_year")] int ReleaseYear,
[property: JsonPropertyName("description")] string Description
);
Complete runnable script
package quickstart;
import java.io.BufferedReader;
import java.io.InputStreamReader;
import java.io.StringReader;
import java.util.List;
import co.elastic.clients.elasticsearch.ElasticsearchClient;
import co.elastic.clients.elasticsearch._types.Refresh;
import co.elastic.clients.elasticsearch.core.BulkRequest;
import co.elastic.clients.elasticsearch.core.SearchResponse;
import co.elastic.clients.elasticsearch.core.search.Hit;
import co.elastic.clients.elasticsearch.esql.EsqlFormat;
import co.elastic.clients.transport.endpoints.BinaryResponse;
public class App {
record Book(
String title,
String author,
int release_year,
String description
) {}
public static void main(String[] args) throws Exception {
ElasticsearchClient es = ElasticsearchClient.of(b -> b
.host(System.getenv("ES_URL"))
.apiKey(System.getenv("ES_API_KEY"))
);
es.indices().create(c -> c
.index("books-complete")
.withJson(new StringReader("""
{
"mappings": {
"properties": {
"description": {
"type": "semantic_text"
}
}
}
}
"""))
);
List<Book> books = List.of(
new Book(
"The Left Hand of Darkness",
"Ursula K. Le Guin",
1969,
"An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap."
),
new Book(
"Project Hail Mary",
"Andy Weir",
2021,
"A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth."
),
new Book(
"The Name of the Wind",
"Patrick Rothfuss",
2007,
"A gifted young musician and magician tells the story of his rise from orphan to legend."
),
new Book(
"Klara and the Sun",
"Kazuo Ishiguro",
2021,
"An artificial friend watches human love and loneliness while hoping a child will pick her."
),
new Book(
"Dune",
"Frank Herbert",
1965,
"On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power."
)
);
BulkRequest.Builder bulk = new BulkRequest.Builder()
.refresh(Refresh.WaitFor);
for (Book book : books) {
bulk.operations(op -> op
.index(idx -> idx
.index("books-complete")
.document(book)
)
);
}
es.bulk(bulk.build());
String semanticQuery = """
{
"query": {
"semantic": {
"field": "description",
"query": "surviving alone in space"
}
}
}
""";
SearchResponse<Book> response = es.search(s -> s
.index("books-complete")
.withJson(new StringReader(semanticQuery)),
Book.class
);
System.out.println("=== Semantic search results ===");
for (Hit<Book> hit : response.hits().hits()) {
System.out.println(" " + hit.score() + " " + hit.source().title());
}
String hybridQuery = """
{
"query": {
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age"
}
}
]
}
}
}
""";
response = es.search(s -> s
.index("books-complete")
.withJson(new StringReader(hybridQuery)),
Book.class
);
System.out.println("\n=== Hybrid search results ===");
for (Hit<Book> hit : response.hits().hits()) {
System.out.println(" " + hit.score() + " " + hit.source().title());
}
String esql = """
FROM "books-complete"
| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)
| KEEP decade, books
| SORT decade ASC
""";
BinaryResponse aggregation = es.esql().query(q -> q
.query(esql)
.format(EsqlFormat.Csv)
);
System.out.println("\n=== Books by decade ===");
try (BufferedReader reader = new BufferedReader(
new InputStreamReader(aggregation.content())
)) {
reader.lines().skip(1).forEach(line -> {
String[] values = line.split(",", -1);
System.out.println(" " + values[0] + "s: " + values[1]);
});
}
es.close();
}
}
Complete runnable script
package main
import (
"bytes"
"context"
"encoding/csv"
"encoding/json"
"fmt"
"log"
"os"
"strings"
"github.com/elastic/go-elasticsearch/v9"
"github.com/elastic/go-elasticsearch/v9/esutil"
"github.com/elastic/go-elasticsearch/v9/typedapi/types/enums/esqlformat"
)
type book struct {
Title string `json:"title"`
Author string `json:"author"`
ReleaseYear int `json:"release_year"`
Description string `json:"description"`
}
func printSearch(es *elasticsearch.Client, query string) {
res, err := es.Search(
es.Search.WithContext(context.Background()),
es.Search.WithIndex("books-complete"),
es.Search.WithBody(strings.NewReader(query)),
)
if err != nil {
log.Fatal(err)
}
defer res.Body.Close()
if res.IsError() {
log.Fatal(res)
}
var response struct {
Hits struct {
Hits []struct {
Score float64 `json:"_score"`
Source book `json:"_source"`
} `json:"hits"`
} `json:"hits"`
}
if err := json.NewDecoder(res.Body).Decode(&response); err != nil {
log.Fatal(err)
}
for _, hit := range response.Hits.Hits {
fmt.Println(" ", hit.Score, hit.Source.Title)
}
}
func main() {
es, err := elasticsearch.New(
elasticsearch.WithAddresses(os.Getenv("ES_URL")),
elasticsearch.WithAPIKey(os.Getenv("ES_API_KEY")),
)
if err != nil {
log.Fatal(err)
}
mapping := strings.NewReader(`{
"mappings": {
"properties": {
"description": {
"type": "semantic_text"
}
}
}
}`)
res, err := es.Indices.Create(
"books-complete",
es.Indices.Create.WithBody(mapping),
)
if err != nil {
log.Fatal(err)
}
if res.IsError() {
log.Fatal(res)
}
res.Body.Close()
books := []book{
{
Title: "The Left Hand of Darkness", Author: "Ursula K. Le Guin",
ReleaseYear: 1969,
Description: "An envoy visits an icy planet whose people have no fixed gender, feeling out politics and friendship across a deep cultural gap.",
},
{
Title: "Project Hail Mary", Author: "Andy Weir",
ReleaseYear: 2021,
Description: "A lone astronaut wakes with amnesia on a spaceship and has to stop a disaster that threatens all life on Earth.",
},
{
Title: "The Name of the Wind", Author: "Patrick Rothfuss",
ReleaseYear: 2007,
Description: "A gifted young musician and magician tells the story of his rise from orphan to legend.",
},
{
Title: "Klara and the Sun", Author: "Kazuo Ishiguro",
ReleaseYear: 2021,
Description: "An artificial friend watches human love and loneliness while hoping a child will pick her.",
},
{
Title: "Dune", Author: "Frank Herbert",
ReleaseYear: 1965,
Description: "On a desert planet prized for a rare spice, a young heir is pulled into a war over ecology, religion, and power.",
},
}
indexer, err := esutil.NewBulkIndexer(esutil.BulkIndexerConfig{
Client: es,
Index: "books-complete",
Refresh: "wait_for",
})
if err != nil {
log.Fatal(err)
}
for _, item := range books {
body, err := json.Marshal(item)
if err != nil {
log.Fatal(err)
}
if err := indexer.Add(context.Background(), esutil.BulkIndexerItem{
Action: "index",
Body: bytes.NewReader(body),
}); err != nil {
log.Fatal(err)
}
}
if err := indexer.Close(context.Background()); err != nil {
log.Fatal(err)
}
fmt.Println("=== Semantic search results ===")
printSearch(es, `{
"query": {
"semantic": {
"field": "description",
"query": "surviving alone in space"
}
}
}`)
fmt.Println("\n=== Hybrid search results ===")
printSearch(es, `{
"query": {
"bool": {
"should": [
{
"match": {
"title": "wind"
}
},
{
"semantic": {
"field": "description",
"query": "young magician coming of age"
}
}
]
}
}
}`)
typed, err := elasticsearch.NewTyped(
elasticsearch.WithAddresses(os.Getenv("ES_URL")),
elasticsearch.WithAPIKey(os.Getenv("ES_API_KEY")),
)
if err != nil {
log.Fatal(err)
}
query := "FROM \"books-complete\"\n" +
"\t\t| STATS books = COUNT(*) BY decade = release_year - (release_year % 10)\n" +
"\t\t| KEEP decade, books\n" +
"\t\t| SORT decade ASC"
aggregation, err := typed.Esql.Query().
Query(query).
Format(esqlformat.Csv).
Do(context.Background())
if err != nil {
log.Fatal(err)
}
rows, err := csv.NewReader(bytes.NewReader(aggregation)).ReadAll()
if err != nil {
log.Fatal(err)
}
fmt.Println("\n=== Books by decade ===")
for _, row := range rows[1:] {
fmt.Printf(" %ss: %s\n", row[0], row[1])
}
}
If you want to continue using your Vector Database project, select a next step based on your goal.
- Vector search: Learn about vectors, embeddings, vector fields, and similarity search in Elasticsearch.
- Compare Elastic Inference Service models: Learn about the embedding models available through the Elastic Inference Service and compare their capabilities.
- Learn about chunking strategies: Learn how to split long text for inference using sentence, word, recursive, or custom chunking.
- Tune retrieval with ranking and reranking: Learn how to improve result ordering with multi-stage ranking and reranking.
- Plan quantization and sizing: Learn how quantization reduces vector memory and storage requirements and compare the available quantization types.
- Build multimodal search in Elasticsearch: Learn how to index images and search them with text, image, and PDF input.
To learn how continued usage is billed, refer to Serverless project billing dimensions. If you don't plan to continue using the project, delete it to avoid additional usage charges.