注: Vertex AI Search は Agent Search に名称変更されます。現在、新しいブランディングを反映するようにコンテンツの更新を進めています。

回答とフォローアップを取得する

このページでは、Agent Search の回答とフォローアップによる検索について説明します。また、メソッド呼び出しを使用してカスタム検索アプリに実装する方法についても説明します。

回答とフォローアップによる検索は、answer メソッドに基づいています。answer メソッドは、古い search メソッドの要約機能と、非推奨の converse メソッドのすべての機能を置き換えます。 answer メソッドには、複雑なクエリを処理する機能など、重要な追加機能もあります。

answer メソッドの機能

answer メソッドの主な機能は次のとおりです。

複雑なクエリに対する回答を生成する機能。たとえば、answer メソッドでは、次のような複合クエリを複数の小さなクエリに分割して、より良い結果を返すことができます。これにより、より良い回答が得られます。
- 「2024 年の Google Cloud と Google 広告のそれぞれの収益は？」
- 「Google は創業から何年後に 10 億米ドルの収益を達成しましたか？」
各ターンで answer メソッドを呼び出して、マルチターンの会話で検索と回答の生成を組み合わせる機能。
search メソッドと組み合わせて検索レイテンシを短縮する機能。search メソッドと answer メソッドを別々に呼び出し、検索結果と回答を異なる iframe に異なるタイミングでレンダリングできます。つまり、ユーザーに検索結果（10 個の青いリンク）をミリ秒単位で表示できます。検索結果を表示する前に、回答が生成されるのを待つ必要はありません。

回答とフォローアップの機能は、クエリ、検索、回答の 3 つのフェーズに分けることができます。

クエリフェーズの機能
検索フェーズの機能
回答フェーズの機能

回答と検索を使用するタイミング

Agent Search には、アプリのクエリに使用される 2 つのメソッドがあります。両者には異なる機能がありますが、重複する機能もあります。

answer メソッドは、次の場合に使用します。

検索結果の AI 生成の回答（またはサマリー）が必要な場合。
マルチターンの検索（コンテキストを保持してフォローアップの質問を可能にする検索）が必要な場合。

search メソッドは、次の場合に使用します。

生成された回答ではなく、検索結果のみが必要。
次のいずれかがある場合:
- メディアまたは医療データ
- 独自のエンベディング
- 類義語またはリダイレクトの制御
- ファセット
- ユーザーの国コード
汎用データストア内のデータを閲覧する必要があります。

次の場合は、answer メソッドと search メソッドを併用します。

10 件を超える検索結果を返す必要があり、かつ、生成された回答が必要な場合。
レイテンシの問題があり、生成された回答が返される前に検索結果を迅速に返して表示したい場合。

クエリフェーズの機能

回答とフォローアップ機能は、自然言語クエリ処理をサポートしています。

このセクションでは、クエリの言い換えと分類のさまざまなオプションについて説明します。

クエリの言い換え

クエリの言い換えはデフォルトでオンになっています。この機能は、検索結果を改善するために、クエリを言い換える最適な方法を自動的に選択します。この機能は、言い換えを必要としないクエリにも対応できます。

複雑なクエリを複数のクエリに分割し、同期サブクエリを実行します。

たとえば、複雑なクエリを 4 つの小さく単純なクエリに分割します。

ユーザー入力	複雑なクエリから作成されたサブクエリ
Andie Ram と Arnaud Clément は、どのような仕事と趣味が共通していますか？	Andie Ram の職業 Arnaud Clément の職業 Andie Ram の趣味 Arnaud Clément の趣味

マルチターンのクエリを合成して、フォローアップの質問をコンテキスト認識型でステートフルにします。

例: 各ターンのユーザー入力から合成されたクエリは次のようになります。

ユーザー入力	合成されたクエリ
ターン 1: 学校用ノートパソコン	学校用ノートパソコン
ターン 2: mac 以外	mac 以外の学校用ノートパソコン
ターン 3: 大画面かつワイヤレスキーボードとマウスも必要	ワイアレスキーボードとマウス付きの、mac 以外の大画面の学校用ノートパソコン
ターン 4: それにパソコン用のバックパック	ワイアレスキーボードとマウスとバックパック付きの、mac 以外の大画面の学校用ノートパソコン

長いクエリを簡素化して、取得を改善します（生成された回答オプションが必要です）。

たとえば、長いクエリを一般的なクエリにします。

ユーザー入力	簡素化されたクエリ
ウェブサイトの [カートに追加] ボタンが正しく機能しない理由を調べています。ユーザーがボタンをクリックしても、商品がカートに追加されず、エラーメッセージが表示されるようです。コードを確認しましたが、正しいようです。問題が何なのかわかりません。この問題のトラブルシューティングを手伝っていただけますか？	ウェブサイトで [カートに追加] ボタンが機能しません。

ユーザー入力

簡素化されたクエリ

ウェブサイトの [カートに追加] ボタンが正しく機能しない理由を調べています。ユーザーがボタンをクリックしても、商品がカートに追加されず、エラーメッセージが表示されるようです。コードを確認しましたが、正しいようです。問題が何なのかわかりません。この問題のトラブルシューティングを手伝っていただけますか？

ウェブサイトで [カートに追加] ボタンが機能しません。

マルチステップの推論を実行する

マルチステップ推論は ReACT（理由 + 行動）パラダイムに基づいており、LLM が自然言語推論を使用して複雑なタスクを解決できるようにします。デフォルトでは、最大ステップ数は 5 です。

次に例を示します。

ユーザー入力	回答を生成する 2 つのステップ
Google は創業から何年後に 10 億米ドルの収益を達成しましたか？	ステップ 1: [思考]: Google の創立日を知る必要があります。そうすれば、それ以降の収益をクエリできます。 [行動] 検索: Google はいつ設立されましたか？[検索結果の確認]:「1998 年」ステップ 2: [思考]: 次は、1998 年以降の Google の年間収益を調べて、初めて 10 億ドルを超えたのはいつかを確認する必要があります。 [行動] 検索: 1998 年以降の Google の収益 [検索結果の確認] 1998 年の Google の収益、1999 年の Google の収益 [回答]: Google は、2003 年に 10 億米ドル以上の収益を達成しました [1]、1998 年の創業から 5 年後です [2]。

ユーザー入力

回答を生成する 2 つのステップ

Google は創業から何年後に 10 億米ドルの収益を達成しましたか？

ステップ 1:
[思考]: Google の創立日を知る必要があります。そうすれば、それ以降の収益をクエリできます。
[行動] 検索: Google はいつ設立されましたか？[検索結果の確認]:「1998 年」

ステップ 2:
[思考]: 次は、1998 年以降の Google の年間収益を調べて、初めて 10 億ドルを超えたのはいつかを確認する必要があります。
[行動] 検索: 1998 年以降の Google の収益
[検索結果の確認] 1998 年の Google の収益、1999 年の Google の収益
[回答]: Google は、2003 年に 10 億米ドル以上の収益を達成しました [1]、1998 年の創業から 5 年後です [2]。

マルチステップの推論では、[生成された回答] オプションを有効にする必要があります。

クエリ分類

クエリ分類オプションは、敵対的なクエリと回答を求めていないクエリを特定します。デフォルトでは、クエリ分類オプションはオフになっています。

敵対的クエリと回答を求めていないクエリの詳細については、敵対的クエリを無視するとサマリーを求めていないクエリを無視するをご覧ください。

検索フェーズの機能

検索の場合、answer メソッドには search メソッドと同じオプションがあります。次に例を示します。

フィルタを適用して、検索を特定のドキュメントに限定する。詳細については、構造化データまたは非構造化データのカスタム検索をフィルタするをご覧ください。
セーフサーチを適用して、暴力やポルノなどの露骨な表現を含むコンテンツを除外します。詳細については、エージェント検索の安全設定をご覧ください。
検索から返されたドキュメントの順位を上げる、または下げるブースト条件を指定する。詳細については、検索結果をブーストするをご覧ください。

回答フェーズの機能

回答フェーズでは、検索結果から回答が生成されたときに、search メソッドと同じ機能を有効にできます。次に例を示します。

回答内の各文のソースを示す引用を取得します。詳しくは、引用を含めるをご覧ください。
プロンプトのプリアンブルを使用して、回答のトーンとスタイル、詳細度などをカスタマイズします。詳細については、カスタムプリアンブルを指定するをご覧ください。
回答の生成に使用する Vertex AI モデルを選択します。詳細については、回答生成モデルのバージョンとライフサイクルをご覧ください。
敵対的または回答を求めていないと分類されたクエリを無視するかどうかを選択します。

敵対的クエリと回答を求めていないクエリの詳細については、敵対的クエリを無視するとサマリーを求めていないクエリを無視するをご覧ください。回答を求めていないクエリは、サマリーを求めていないクエリとも呼ばれます。

search メソッドでは利用できない回答フェーズの機能は次のとおりです。

各クレーム（生成された回答内の文）のサポートスコアの取得。サポートスコアは、[0,1] の範囲の浮動小数点値で、主張がデータストア内のデータにおいてどの程度根拠があるかを示します。詳細については、グラウンディングサポートスコアを返すをご覧ください。
回答の集計サポートスコアの取得。サポートスコアは、回答がデータストア内のデータにおいてどの程度根拠があるかを示します。詳細については、グラウンディングサポートスコアを返すをご覧ください。
根拠のある回答のみを返します。特定のサポートスコアのしきい値を満たす回答のみを返すように選択できます。詳細については、根拠のある回答のみを表示するをご覧ください。
関連する質問を返すように選択します。関連する質問は、ユーザーが独自の質問を入力する代わりに選択できる候補です。
クエリにパーソナライズ情報を追加して、個々のユーザーに合わせて回答をカスタマイズできるようにします。詳細については、回答をカスタマイズするをご覧ください。

テキストに加えてグラフや画像を含むマルチモーダル回答を受け取るには、次のオプションを使用できます。

回答に含まれるデータをプロットしたグラフやグラフを含む回答を取得する。詳細については、回答のグラフを生成するをご覧ください。
データストアから画像を取得する。データストアに画像が含まれている場合、回答メソッドは回答で画像を返すことができます。引用がリクエストされた場合、データストアの画像が参照として返されることもあります。詳細については、データストアから既存のイメージを取得するをご覧ください。

始める前に

アプリの種類に応じて、次の要件を完了します。

構造化データ、非構造化データ、ウェブサイトの検索アプリを使用している場合は、次のオプションをオンにします。
- Enterprise エディションの機能: これにより、コアの生成回答機能にアクセスできます。これには、関連する質問、クエリの簡素化、複数ターンのクエリ、画像やグラフを返すマルチモーダル回答などの高度な生成回答機能を除く、すべての回答生成機能が含まれます。
- 生成レスポンス: 複数ステップの推論、クエリの簡素化、マルチターンクエリ、関連する質問、画像やグラフを返すマルチモーダル回答を必要とする高度な生成回答機能を利用できます。
また、ウェブサイト検索データストアがある場合は、[ウェブサイトの高度なインデックス登録] をオンにします。

検索と回答（基本）

次のコマンドは、answer メソッドを呼び出して、生成された回答と、ソースへのリンクを含む検索結果のリストを返す方法を示しています。

このコマンドは、必要な入力のみを表示します。オプションはデフォルトのままにします。

REST

検索して生成された回答付きの結果を取得する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"}
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。たとえば、「BigQuery と Spanner のデータベースを比較してください」などです。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{"query": { "text": "Which database is faster, bigquery or spanner?"}}'
        
{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "The provided sources do not directly compare the speed of BigQuery and Spanner.  However, they do highlight the performance capabilities of each database. BigQuery is described as having strong query performance, particularly for short and complex queries.  It also offers a serverless architecture that provides consistent performance regardless of query complexity.  Spanner is described as having high performance at virtually unlimited scale, with single-digit millisecond latency for strongly-consistent reads and writes.  It also offers a five-nines availability SLA.  Ultimately, the best database for a particular use case will depend on the specific requirements of the application. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": " What is the performance of BigQuery?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/9ab3ef91bcfde1fcd091efe9df7c699c",
                  "uri": "https://cloud.google.com/bigquery/docs/best-practices-performance-overview",
                  "title": "Introduction to optimizing query performance | BigQuery | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "After a query begins execution, \u003cb\u003eBigQuery\u003c/b\u003e calculates how many slots each query stage uses based on the stage size and complexity and the number of slots ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/4e545c5cb69b06b251265114d9099cb4",
                  "uri": "https://cloud.google.com/bigquery/docs/query-insights",
                  "title": "Get query performance insights | BigQuery | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "This document describes how to use the query execution graph to diagnose query \u003cb\u003eperformance\u003c/b\u003e issues, and to see query \u003cb\u003eperformance\u003c/b\u003e insights. \u003cb\u003eBigQuery\u003c/b\u003e offers ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d34672d877eefe596f9c7d1a3d7076b1",
                  "uri": "https://cloud.google.com/bigquery/docs/best-practices-performance-compute",
                  "title": "Optimize query computation | BigQuery | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "After addressing the query \u003cb\u003eperformance\u003c/b\u003e insights, you can further optimize your query by performing the following tasks: Reduce data that is to be processed ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/75ce2f05833683e60ddc21a11ce0466f",
                  "uri": "https://cloud.google.com/blog/products/data-analytics/troubleshoot-and-optimize-your-bigquery-analytics-queries-with-query-execution-graph/",
                  "title": "Troubleshoot and optimize your BigQuery analytics queries with query execution graph | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "Since query \u003cb\u003eperformance\u003c/b\u003e is multi-faceted, \u003cb\u003eperformance\u003c/b\u003e insights might only provide a partial picture of the overall query \u003cb\u003eperformance\u003c/b\u003e. Execution graph. When ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          },
          {
            "searchAction": {
              "query": " What is the performance of Spanner?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f3d036b60379873acf7c73081c5e5b5c",
                  "uri": "https://cloud.google.com/spanner/docs/performance",
                  "title": "Performance overview | Spanner | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "These \u003cb\u003eperformance\u003c/b\u003e improvements should result in higher throughput and better latency in \u003cb\u003eSpanner\u003c/b\u003e nodes in both regional and multi-region instance configurations.",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/422496248ade354c73b4c906b8eb9b5f",
                  "uri": "https://cloud.google.com/blog/products/databases/announcing-cloud-spanner-price-performance-updates",
                  "title": "Announcing Cloud Spanner price-performance updates | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "Alongside lower costs, Cloud \u003cb\u003eSpanner\u003c/b\u003e provides single-digit ms latencies and strong consistency across multiple availability zones in the same region.",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/53c2a1a6990480ba4aa05cc6b4404562",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/understanding-cloud-spanner-performance-metrics-scale-key-visualizer",
                  "title": "Understanding Cloud Spanner performance metrics at scale with Key Visualizer | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "Designed for \u003cb\u003eperformance\u003c/b\u003e tuning and instance sizing, you can use Key Visualizer today in the web-based Cloud Console for all \u003cb\u003eSpanner\u003c/b\u003e databases at no additional ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/a6501ecd5d6391e3ade49097bab0ad3a",
                  "uri": "https://cloud.google.com/blog/products/databases/a-technical-overview-of-cloud-spanners-query-optimizer",
                  "title": "A technical overview of Cloud Spanner's query optimizer | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "... performance. Typically, a join will ... Google is continuously improving out-of-the-box \u003cb\u003eperformance of Spanner\u003c/b\u003e and reducing the need for manual tuning.",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "answerQueryToken": "NMwKDAj1_d62BhC72_X_AhIkNjZkN2I4YWEtMDAwMC0yYTdiLWIxMmMtMDg5ZTA4MjhlNzY0"
}

この例では、クエリは「Spanner のパフォーマンスは？」と「BigQuery のパフォーマンスは？」という部分に分解されています。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

Agent Search に対する認証を行うには、アプリケーションのデフォルト認証情報を設定します。詳細については、ローカル開発環境の認証の設定をご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

クエリフェーズのコマンド

このセクションでは、answer メソッド呼び出しのクエリフェーズにオプションを指定する方法について説明します。

検索と回答（言い換えが無効）
検索と回答（最大ステップ数を指定）
クエリ分類による検索と回答

検索と回答（言い換えが無効）

次のコマンドは、answer メソッドを呼び出して、生成された回答と検索結果のリストを返す方法を示しています。言い換えオプションが無効になっているため、回答が前の回答とは異なる場合があります。

REST

クエリの言い換えを適用せずに検索して、生成された回答を含む結果を取得する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "queryUnderstandingSpec": {
           "queryRephraserSpec": {
              "disable": true
        }
    }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。たとえば、「BigQuery と Spanner のデータベースを比較してください」などです。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{
      "query": { "text": "Which database is faster, bigquery or spanner?"},
      "queryUnderstandingSpec": { "queryRephraserSpec": { "disable": true } }
    }'
    
    {
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "The sources provided do not directly compare the speed of BigQuery and Spanner. They do mention that Spanner is optimized for transactional workloads and BigQuery is optimized for analytical workloads. Spanner is a fully managed relational database that provides seamless replication across regions in Google Cloud. BigQuery is a serverless, highly scalable, and cost-effective multi-cloud data warehouse. Spanner is designed to scale horizontally across multiple regions and continents. BigQuery is designed for business agility. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Which database is faster, bigquery or spanner?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/ecc0e7547253f4ca3ff3328ce89995af",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/how-spanner-and-bigquery-work-together-handle-transactional-and-analytical-workloads",
                  "title": "How Spanner and BigQuery work together to handle transactional and analytical workloads | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "A federated \u003cb\u003equery\u003c/b\u003e might not be as \u003cb\u003efast\u003c/b\u003e as querying local \u003cb\u003eBigQuery tables\u003c/b\u003e. There may be higher latency because of the small wait time for the source \u003cb\u003edatabase\u003c/b\u003e to ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d7e238f73608a860e00b752ef80e2941",
                  "uri": "https://cloud.google.com/blog/products/databases/cloud-spanner-gets-stronger-with-bigquery-federated-queries",
                  "title": "Cloud Spanner gets stronger with BigQuery-federated queries | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "... \u003cb\u003equick\u003c/b\u003e lookup on \u003cb\u003edata\u003c/b\u003e that's in \u003cb\u003eSpanner\u003c/b\u003e -- you can ... Set up an external \u003cb\u003edata\u003c/b\u003e source for the \u003cb\u003eSpanner\u003c/b\u003e shopping \u003cb\u003edatabase\u003c/b\u003e in \u003cb\u003eBigQuery\u003c/b\u003e. ... The \u003cb\u003equery\u003c/b\u003e is executed in ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f3d036b60379873acf7c73081c5e5b5c",
                  "uri": "https://cloud.google.com/spanner/docs/performance",
                  "title": "Performance overview | Spanner | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "The information on this page applies to both GoogleSQL and PostgreSQL \u003cb\u003edatabases\u003c/b\u003e. Note: We are in the process of rolling out \u003cb\u003eperformance\u003c/b\u003e and storage changes that ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/47b09cb5ad5e3ab3b1eb93d99ecb0896",
                  "uri": "https://cloud.google.com/blog/products/databases/rewe-uses-cloud-spanner-to-optimize-for-speed-and-performance",
                  "title": "REWE uses Cloud Spanner to optimize for speed and performance | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "As a fully managed relational \u003cb\u003edatabase\u003c/b\u003e, \u003cb\u003eSpanner\u003c/b\u003e provides unlimited scale, strong consistency, and up to 99.999% availability. By choosing this approach to ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "answerQueryToken": "M8gKCwjp_t62BhC7wOFMEiQ2NmQ3YjhhZS0wMDAwLTJhN2ItYjEyYy0wODllMDgyOGU3NjQ"
}

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

検索と回答（最大ステップ数を指定）

次のコマンドは、answer メソッドを呼び出して、生成された回答と検索結果のリストを返す方法を示しています。言い換えステップの数が増えているため、回答が前の回答とは異なっています。

REST

最大 5 回の言い換えステップを許可して、生成された回答付きで検索し、結果を取得する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "queryUnderstandingSpec": {
            "queryRephraserSpec": {
                "maxRephraseSteps": MAX_REPHRASE
             }
         }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。たとえば、「BigQuery と Spanner のデータベースを比較してください」などです。
MAX_REPHRASE: 言い換えステップの最大数。最大値は 5 です。設定されていない場合、または 1 未満に設定されている場合、値はデフォルトの 1 になります。

コマンドの例

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "How much longer does it take to train a recommendations model than a search model"},
        "queryUnderstandingSpec": {
            "queryRephraserSpec": {
            "maxRephraseSteps": 5
             }
         }
      }'

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

クエリ分類による検索と回答

次のコマンドは、answer メソッドを呼び出して、クエリが敵対的、回答を求めていない、またはどちらでもないかを問い合わせる方法を示しています。

回答にはクエリの分類タイプが含まれますが、回答自体は分類の影響を受けません。クエリタイプに応じて回答の動作を変更する場合は、回答フェーズで変更できます。敵対的クエリを無視するとサマリーを求めていないクエリを無視するをご覧ください。

REST

クエリが敵対的なものか、回答を求めていないものかを判断するには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "queryUnderstandingSpec": {
            "queryClassificationSpec": {
                "types": ["QUERY_CLASSIFICATION_TYPE"]
             }
         }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。例: 「hello」。
QUERY_CLASSIFICATION_TYPE: 識別するクエリタイプ（ADVERSARIAL_QUERY、NON_ANSWER_SEEKING_QUERY、または両方）。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{
      "query": {
          "text": "Hello!"},
          "queryUnderstandingSpec": {
             "queryClassificationSpec": {
             "types": ["ADVERSARIAL_QUERY", "NON_ANSWER_SEEKING_QUERY"]
              }
          }
}'

{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "A user reported that their Google Voice account was randomly sending \"Hello!\" replies to incoming texts. The user was frustrated because they did not want to send these replies and found the behavior random. The user was unable to find any linked accounts, Google extensions, or other settings that could be causing the issue. The user confirmed that Google Voice does not have auto-reply functions. The user was seeking help to stop the automatic replies. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Hello!"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/69e92e5b1de5b1e583fbe95f94dd4cbf",
                  "uri": "https://support.google.com/voice/thread/152245405/google-voice-is-randomly-automatically-sending-hello-replies-to-incoming-texts?hl=en",
                  "title": "Google voice is randomly/automatically sending \"Hello!\" replies to incoming texts",
                  "snippetInfo": [
                    {
                      "snippet": "There IS a new "Smart reply" feature on the Android or iOS client apps, but you'd have to a) receive a SMS/MMS, b) open it up, c) look at the three suggested ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/44fb313bcc09877e7239f3810ddb132b",
                  "uri": "https://support.google.com/mail/thread/58174131/gmail-sends-random-email-saying-hello-to-my-emails-without-me-touching-it?hl=en",
                  "title": "Gmail sends random email saying \"Hello!!\" to my emails without me touching it",
                  "snippetInfo": [
                    {
                      "snippet": "Gmail sends random email saying "\u003cb\u003eHello\u003c/b\u003e!!" to my emails without me touching it. Whenever I email somebody and they reply, a random email from my Gmail is sent to ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/16d65e2af7fa854d1a00995525646dc3",
                  "uri": "https://support.google.com/voice/thread/112990484/google-voice-sending-hello-in-response-to-text-messages?hl=en",
                  "title": "Google Voice sending \"Hello,\" in response to text messages",
                  "snippetInfo": [
                    {
                      "snippet": "When I receive text messages, a reply is instantly sent out reading "\u003cb\u003eHello\u003c/b\u003e," and I cannot figure out how this is happening. I have no linked accounts, ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/a828eb8f442f1dfbdda06dbeb52841b0",
                  "uri": "https://support.google.com/a/thread/161821861/hello-hello-the-lost-phone?hl=en",
                  "title": "Hello.Hello the lost phone - Google Workspace Admin Community",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eHello\u003c/b\u003e the lost phone. My wife lost her phone but she cannot remember her emails pasward to help track .",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ],
    "queryUnderstandingInfo": {
      "queryClassificationInfo": [
        {
          "type": "ADVERSARIAL_QUERY"
        },
        {
        "type": "NON_ANSWER_SEEKING_QUERY",
          "positive": true
        }
      ]
    }
  },
  "answerQueryToken": "NMwKDAjVloK3BhCdt8u9AhIkNjZkYmFhNWItMDAwMC0yZTBkLTg0ZDAtMDg5ZTA4MmRjYjg0"
}

この例では、クエリ「hello」は敵対的ではありませんが、回答を求めていないものとして分類されます。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

検索フェーズのコマンド: 検索して検索結果のオプション付きで回答する

このセクションでは、answer メソッド呼び出しの検索フェーズ部分のオプション（返されるドキュメントの最大数、ブースト処理、フィルタリングなどのオプション）を指定する方法と、独自の検索結果を指定して回答を取得する方法について説明します。

次のコマンドは、answer メソッドを呼び出して、検索結果を返す方法に関するさまざまなオプションを指定する方法を示しています。（検索結果は回答とは関係ありません）

REST

検索結果が返される内容と方法に関連するさまざまなオプションを設定するには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
          "searchSpec": {
          "searchParams": {
            "maxReturnResults": MAX_RETURN_RESULTS,
            "filter": "FILTER",
            "boostSpec": BOOST_SPEC,
            "orderBy": "ORDER_BY",
            "searchResultMode": SEARCH_RESULT_MODE
           }
         }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。たとえば、「BigQuery データベースと Spanner データベースを比較してください」などです。
MAX_RETURN_RESULTS: 返される結果の数。デフォルト値は 10 です。最大値は 25 です。
FILTER: フィルタは、クエリ対象のドキュメントを指定します。ドキュメントのメタデータがフィルタ仕様を満たしている場合、ドキュメントはクエリされます。フィルタ構文などの詳細については、構造化データまたは非構造化データのカスタム検索をフィルタするをご覧ください。
BOOST_SPEC: ブースト仕様を使用すると、検索結果で特定のドキュメントをブーストできます。これにより、回答に影響する可能性があります。ブースト仕様の構文など、詳細については、検索結果をブーストするをご覧ください。
ORDER_BY: ドキュメントが返される順序。ドキュメントは、Document オブジェクトのフィールドで並べ替えることができます。orderBy 式では大文字と小文字が区別されます。このフィールドが認識できない場合、INVALID_ARGUMENT が返されます。
SEARCH_RESULT_MODE: 検索結果モード（DOCUMENTS または CHUNKS）を指定します。詳細については、ドキュメントを解析してチャンクすると ContentSearchSpec をご覧ください。このフィールドは、API の v1alpha バージョンでのみ使用できます。

コマンドの例と結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{
      "query": {
          "text": "Does spanner database have an API?"},
          "searchSpec": {
          "searchParams": { "maxReturnResults": 3 }
          }
    }'
    
{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "Spanner database has an API that provides programmatic access to the database. The API is available through client libraries, RPC, and REST. The client libraries allow you to interact with Spanner in your preferred language. The RPC API and REST API provide programmatic access to Spanner. The Cloud Spanner API is a managed, mission-critical, globally consistent and scalable relational database service. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Does spanner database have an API?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d135b46c4a44d0cc6b652538c1887f4d",
                  "uri": "https://cloud.google.com/spanner/docs/apis",
                  "title": "APIs & reference | Spanner | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "The client libraries, the RPC \u003cb\u003eAPI\u003c/b\u003e, and the REST \u003cb\u003eAPI\u003c/b\u003e provide programmatic access to \u003cb\u003eSpanner\u003c/b\u003e. \u003cb\u003eSpanner\u003c/b\u003e client libraries. \u003cb\u003eGet\u003c/b\u003e started with \u003cb\u003eSpanner\u003c/b\u003e in your language ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/7a744d43e61ccd33539de74d5c1f6313",
                  "uri": "https://cloud.google.com/spanner/docs/reference/rest",
                  "title": "Cloud Spanner API",
                  "snippetInfo": [
                    {
                      "snippet": "Returns permissions that the caller \u003cb\u003ehas\u003c/b\u003e on the specified \u003cb\u003edatabase\u003c/b\u003e or backup resource. updateDdl, PATCH /v1/{\u003cb\u003edatabase\u003c/b\u003e=projects/*/instances/*/\u003cb\u003edatabases\u003c/b\u003e/*}/ddl",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/70834ebf4b72b6dc69e06c44ee80f90b",
                  "uri": "https://cloud.google.com/spanner/docs/reference/rpc",
                  "title": "Cloud Spanner API",
                  "snippetInfo": [
                    {
                      "snippet": "ChangeQuorum \u003cb\u003eis\u003c/b\u003e strictly restricted to \u003cb\u003edatabases\u003c/b\u003e ... Returns the schema of a Cloud \u003cb\u003eSpanner database\u003c/b\u003e ... Returns permissions that the caller \u003cb\u003ehas\u003c/b\u003e on the specified ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "answerQueryToken": "NMwKDAj2l4K3BhCqiv66ARIkNjZkYmFhNmMtMDAwMC0yZTBkLTg0ZDAtMDg5ZTA4MmRjYjg0"
}

この例では、返されるドキュメントの数は 3 つに制限されます。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

回答フェーズのコマンド

このセクションでは、answer メソッド呼び出しをカスタマイズする方法について説明します。必要に応じて、次のオプションを組み合わせることができます。

敵対的クエリと回答を求めていないクエリを無視する
関連する回答のみを表示する
グラウンディングサポートスコアを返す
根拠のある回答のみを表示する
回答モデルを指定する
カスタムプリアンブルを指定する
引用を含める
回答の言語コードを設定する
回答をパーソナライズする
回答のグラフを生成する
データストアから既存の画像を取得する

敵対的クエリと回答を求めていないクエリを無視する

次のコマンドは、answer メソッドを呼び出すときに、敵対的クエリや回答を求めていないクエリに回答しないようにする方法を示しています。

REST

敵対的または回答を求めていないクエリへの回答をスキップするには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
           "ignoreAdversarialQuery": true,
           "ignoreNonAnswerSeekingQuery": true
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)"
    -H "Content-Type: application/json"
    "https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
    -d '{
      "query": { "text": "Hello"},
      "answerGenerationSpec": {
          "ignoreAdversarialQuery": true,
          "ignoreNonAnswerSeekingQuery": true }
        }'
        
        {
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "A summary could not be generated for your search query. Here are some search results.",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Hello"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/69e92e5b1de5b1e583fbe95f94dd4cbf",
                  "uri": "https://support.google.com/voice/thread/152245405/google-voice-is-randomly-automatically-sending-hello-replies-to-incoming-texts?hl=en",
                  "title": "Google voice is randomly/automatically sending \"Hello!\" replies to incoming texts",
                  "snippetInfo": [
                    {
                      "snippet": "There IS a new "Smart reply" feature on the Android or iOS client apps, but you'd have to a) receive a SMS/MMS, b) open it up, c) look at the three suggested ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/16d65e2af7fa854d1a00995525646dc3",
                  "uri": "https://support.google.com/voice/thread/112990484/google-voice-sending-hello-in-response-to-text-messages?hl=en",
                  "title": "Google Voice sending \"Hello,\" in response to text messages",
                  "snippetInfo": [
                    {
                      "snippet": "When I receive text messages, a reply is instantly sent out reading "\u003cb\u003eHello\u003c/b\u003e," and I cannot figure out how this is happening. I have no linked accounts, ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/b3bdde4957f588a1458c533269626d09",
                  "uri": "https://support.google.com/voice/thread/4307458/lately-an-automatic-text-response-saying-hello-is-going-out-how-do-i-stop-this?hl=en",
                  "title": "Lately an automatic text response saying, \"Hello\" is going out. How do I stop this? - Google Voice Community",
                  "snippetInfo": [
                    {
                      "snippet": "You need to find out what app is causing it and deactivate or delete it. Last edited Apr 16, 2019.",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/a828eb8f442f1dfbdda06dbeb52841b0",
                  "uri": "https://support.google.com/a/thread/161821861/hello-hello-the-lost-phone?hl=en",
                  "title": "Hello.Hello the lost phone - Google Workspace Admin Community",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eHello\u003c/b\u003e the lost phone. My wife lost her phone but she cannot remember her emails pasward to help track .",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ],
    "answerSkippedReasons": [
      "NON_ANSWER_SEEKING_QUERY_IGNORED"
    ]
  },
  "answerQueryToken": "NMwKDAjFgN-2BhDlsKaZARIkNjZkN2I0NmItMDAwMC0yZmQ5LTkwMDktZjQwMzA0M2E5YTg4"
}

この例では、クエリが回答を求めていないと判断されたため、回答は生成されません。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

REST

関連性の高い結果が見つからない場合にフォールバック回答を返すには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
           "ignoreLowRelevantContent": true
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。

コマンドの例と結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{"query": { "text": "foobar"}, "answerGenerationSpec": {
       "ignoreLowRelevantContent": true
    } }'
    
 {
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "We do not have a summary for your query.",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "foobar"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/719b79786f0c143717c569eade5305d9",
                  "uri": "https://support.google.com/websearch/thread/261714267/google-foobar-bug-console-disappeared?hl=en",
                  "title": "Google Foobar Bug - Console Disappeared",
                  "snippetInfo": [
                    {
                      "snippet": "Google \u003cb\u003eFoobar\u003c/b\u003e Bug - Console Disappeared. After I logged in today the top bar says "The \u003cb\u003eFoobar\u003c/b\u003e Challenge will be turned down on 1 April 2024. If you run out of ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/932369826585ff45f6ab3eba01ba6933",
                  "uri": "https://support.google.com/websearch/thread/95251114/unable-to-contact-foobar-recruiter?hl=en",
                  "title": "Unable to contact Foobar Recruiter - Google Search Community",
                  "snippetInfo": [
                    {
                      "snippet": "Access is by invitation only so you will need to have the proper credentials to login. You can always reach out using the contact us button, but there is no ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/fb736a30ff90d058be755f0a04a522a8",
                  "uri": "https://support.google.com/websearch/thread/121151780/foobar-challenge-appeared-to-me-then-disappeared?hl=en",
                  "title": "Foobar challenge appeared to me then disappeared - Google Search Community",
                  "snippetInfo": [
                    {
                      "snippet": "Hi. I got the \u003cb\u003efoobar\u003c/b\u003e challenge some months ago. But then it disappeared immediately, maybe by misclick (though I don't think I misclicked).",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f56f2656b0d02b839509d0e67e60c1c9",
                  "uri": "https://support.google.com/chrome/thread/159931759/cannot-access-google-foobar-challenge?hl=en",
                  "title": "Cannot Access Google FooBar Challenge",
                  "snippetInfo": [
                    {
                      "snippet": "I knew I wouldn't have time for it today, so I just kept the tab in the background. Tonight, I went to go close all my tabs, but the page had changed. It said " ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ],
    "answerSkippedReasons": [
      "NO_RELEVANT_CONTENT"
    ]
  },
  "answerQueryToken": "M8gKCwiokvy2BhDtv8EDEiQ2NmQ5NDQxZC0wMDAwLTIxMGQtOWU2Yi1mNDAzMDQ1ZGJkMzA"
}

この例では、結果がクエリに十分に関連していないと判断されたため、生成された回答と結果の代わりにフォールバック回答が返されました。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

グラウンディングサポートスコアを返す

次のコマンドは、回答と主張に対するグラウンディングサポートスコアを返す方法を示しています。

Vertex AI でのグラウンディングの一般的な情報については、RAG でグラウンディングをチェックするをご覧ください。groundingConfigs.check メソッドは、answer メソッドによって呼び出されます。

REST

各クレーム（回答内の文）のサポートスコアと回答の集計サポートスコアを返すには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "groundingSpec": {
           "includeGroundingSupports": true,
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer"
-d '{
    "query": { "text": "What is SQL?"},
    "groundingSpec": {
       "includeGroundingSupports": true,
    }
  }'
            
{
"answer": {
"state": "SUCCEEDED",
"answerText": "SQL stands for Structured Query Language. It is a database management programming language that is used to access and manage data in a database. SQL is used to create, update, and delete data in a database. It can also be used to query data and retrieve information. SQL is a standard language that is used by many different database systems.",
"groundingScore" 0.9
"groundingSupports": [
  {
    "endIndex": "41",
    "sources": [
      {
        "referenceId": "1"
      }
    ]
    "groundingScore": 0.9
    "groundingCheckRequired": true
  },
  {
    "startIndex": "42",
    "endIndex": "144",
    "sources": [
      {
        "referenceId": "1"
      }
    ]
    "groundingScore": 0.8
    "groundingCheckRequired": true
  },
  {
    "startIndex": "267",
    "endIndex": "342",
    "sources": [
      {
        "referenceId": "2"
      }
    ]
    "groundingScore": 0.6
    "groundingCheckRequired": true
  }
],
"references": [
  {
    "chunkInfo": {
      "content": "There are a lot of Databases available in the market such as MS Access, Oracle and many others.For you to write programs that interact with these databases easily, there has to be a way where you could get information from all these databases using the same method.For this purpose SQL was developed.It is a kind of language (simple when compared to the likes of C or C++) which enables you to ask all your queries to a database without bothering about the exact type of database.When you use this Query the database engine would first find the table called people.Then it would find a column called firstname.Next it would compare all the values in that column with 'Reena'.Finally it would return all the details wherever it finds a match for the firstname.When you write a database program in VC++ or Java or any other language for that matter, you would make a database connection to your database and then you would query the database using SQL queries.When you query the database with any SQL query the database returns a recordset.A recordset is basically a set of records (all the entries that your query returns).This recordset is received in your program and all languages have a data structure which represents a recordset.",
      "documentMetadata": {
        "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d993d922043374f5ef7ba297c158b106",
        "uri": "gs://my-bucket-123/documents/058dee0ec23a3e92f9bfd7cd29840e8f.txt"
        "structData": {
          "fields": [
            {
              "key": "cdoc_url"
              "value": { "stringValue": "058dee0ec23a3e92f9bfd7cd29840e8f" }
            },
            {
              "key": "doc_id"
              "value": { "stringValue": "d993d922043374f5ef7ba297c158b106" }
            }
          ]
        }
      }
    }
  },
  {
    "chunkInfo": {
      "content": "The Structured Query Language (SQL) is a database management programming language.SQL is a tool for accessing databases, and more specifically, relational databases, and can be used with different database products.This chapter will prepare you to learn basic database management using this language.SQLite – To implement SQL as a library, you need SQLite.SQLite is intended to provide users and programs a way to store data using a SQL interface within the program.SQLite3 can be used to manipulate SQLite databases for major Linux distros.SQL is used to access relational databases.Each database contains more or less tables which in turn contain more or less rows and columns.Hereby a single row is seen as a separate object with features represented by the tables' columns.To access a table's data you first have to connect to its database.With the same table, the query SELECT * FROM T WHERE C1 = 1 will result in all the elements of all the rows where the value of column C1 is '1' being shown.A WHERE clause specifies that a SQL statement should only affect rows that meet specified criteria.The criteria are expressed in the form of predicates.WHERE clauses are not mandatory clauses of SQL statements, but should be used to limit the number of rows affected by a SQL DML statement or returned by a query.",
      "documentMetadata": {
        "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/3825eac51ef9e934bbc558faa42f4c71",
        "uri": "gs://my-bucket-123/documents/26f5872b0719790cb966a697bfa1ea27.txt"
        "structData": {
          "fields": [
            {
              "key": "cdoc_url"
              "value": { "stringValue": "26f5872b0719790cb966a697bfa1ea27" }
            },
            {
              "key": "doc_id"
              "value": { "stringValue": "3825eac51ef9e934bbc558faa42f4c71" }
            }
          ]
        }
      }
    }
  },
  {
    "chunkInfo": {
      "content": "This chapter focuses on using Paradox as a client/server development tool.It does not talk about connecting; it is assumed you have already connected.If you are having trouble connecting to a particular SQL server, then refer to the Connection Guide for that particular server.This chapter does review what a user can do interactively with Paradox and how to use ObjectPAL with SQL servers.Structured Query Language (SQL) was developed to create a standard for accessing database information.The ANSI standard for SQL allows a user to become familiar with the commands needed to query many different types of data.After you learn ANSI SQL, you then can query many different databases.Is SQL a solid standard?Yes and no.Yes, the core ANSI SQL commands are solid and consistent from vendor to vendor.Every vendor, however, adds capability to its version of SQL.These improvements are expected because ANSI SQL does not go far enough to cover every feature of every high-end DBMS.The SQL standard is used by many companies for their high-end products.They include Oracle, Sybase, Microsoft SQL, Informix, and Interbase.Paradox also provides the capability to use standard ANSI SQL commands on local Paradox and dBASE tables.Although SQL by definition is a standard, various flavors are on the market.",
      "documentMetadata": {
        "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/b3e88db8676b87b99af1e6ecc7d8757f",
        "uri": "gs://my-bucket-123/documents/073c21335d37d8d14982cb3437a721c0.txt"
        "structData": {
          "fields": [
            {
              "key": "cdoc_url"
              "value": { "stringValue": "073c21335d37d8d14982cb3437a721c0" }
            },
            {
              "key": "doc_id"
              "value": { "stringValue": "b3e88db8676b87b99af1e6ecc7d8757f" }
            }
          ]
        }
      }
    }
  }
],
...
"steps": [
  {
    "state": "SUCCEEDED",
    "description": "Rephrase the query and search.",
    "actions": [
      {
        "searchAction": {
          "query": "What is SQL?"
        },
        "observation": {
          "searchResults": [
            {
              "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d993d922043374f5ef7ba297c158b106",
              "uri": "gs://my-bucket-123/documents/058dee0ec23a3e92f9bfd7cd29840e8f.txt",
              "chunkInfo": [
                {
                  "content": "There are a lot of Databases available in the market such as MS Access, Oracle and many others.For you to write programs that interact with these databases easily, there has to be a way where you could get information from all these databases using the same method.For this purpose SQL was developed.It is a kind of language (simple when compared to the likes of C or C++) which enables you to ask all your queries to a database without bothering about the exact type of database.When you use this Query the database engine would first find the table called people.Then it would find a column called firstname.Next it would compare all the values in that column with 'Reena'.Finally it would return all the details wherever it finds a match for the firstname.When you write a database program in VC++ or Java or any other language for that matter, you would make a database connection to your database and then you would query the database using SQL queries.When you query the database with any SQL query the database returns a recordset.A recordset is basically a set of records (all the entries that your query returns).This recordset is received in your program and all languages have a data structure which represents a recordset."
                }
              ]
            },
            {
              "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/3825eac51ef9e934bbc558faa42f4c71",
              "uri": "gs://my-bucket-123/documents/26f5872b0719790cb966a697bfa1ea27.txt",
              "chunkInfo": [
                {
                  "content": "The Structured Query Language (SQL) is a database management programming language.SQL is a tool for accessing databases, and more specifically, relational databases, and can be used with different database products.This chapter will prepare you to learn basic database management using this language.SQLite – To implement SQL as a library, you need SQLite.SQLite is intended to provide users and programs a way to store data using a SQL interface within the program.SQLite3 can be used to manipulate SQLite databases for major Linux distros.SQL is used to access relational databases.Each database contains more or less tables which in turn contain more or less rows and columns.Hereby a single row is seen as a separate object with features represented by the tables' columns.To access a table's data you first have to connect to its database.With the same table, the query SELECT * FROM T WHERE C1 = 1 will result in all the elements of all the rows where the value of column C1 is '1' being shown.A WHERE clause specifies that a SQL statement should only affect rows that meet specified criteria.The criteria are expressed in the form of predicates.WHERE clauses are not mandatory clauses of SQL statements, but should be used to limit the number of rows affected by a SQL DML statement or returned by a query."
                }
              ]
            },
            {
              "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/b3e88db8676b87b99af1e6ecc7d8757f",
              "uri": "gs://my-bucket-123/documents/073c21335d37d8d14982cb3437a721c0.txt",
              "chunkInfo": [
                {
                  "content": "This chapter focuses on using Paradox as a client/server development tool.It does not talk about connecting; it is assumed you have already connected.If you are having trouble connecting to a particular SQL server, then refer to the Connection Guide for that particular server.This chapter does review what a user can do interactively with Paradox and how to use ObjectPAL with SQL servers.Structured Query Language (SQL) was developed to create a standard for accessing database information.The ANSI standard for SQL allows a user to become familiar with the commands needed to query many different types of data.After you learn ANSI SQL, you then can query many different databases.Is SQL a solid standard?Yes and no.Yes, the core ANSI SQL commands are solid and consistent from vendor to vendor.Every vendor, however, adds capability to its version of SQL.These improvements are expected because ANSI SQL does not go far enough to cover every feature of every high-end DBMS.The SQL standard is used by many companies for their high-end products.They include Oracle, Sybase, Microsoft SQL, Informix, and Interbase.Paradox also provides the capability to use standard ANSI SQL commands on local Paradox and dBASE tables.Although SQL by definition is a standard, various flavors are on the market."
                }
              ]
            },
...
            {
              "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/3dd4014e41044c5dd6a0fe380847f369",
              "uri": "gs://my-bucket-123/documents/76245cb33a66f4fbd9030a2a11eea00d.txt",
              "chunkInfo": [
                {
                  "content": "SQL injection is a code injection technique that might destroy your database.You can read more here OWASP sql injection testing sheet.Description: SQL injection ( second order) SQL injection vulnerabilities arise when user- controllable data is incorporated sheet into database SQL queries in an unsafe manner.This sheet cheat wiki assumes you have a basic understanding of SQL injection, please go here for an introduction if you are unfamiliar.Bypass login page with sql SQL injection [ closed].Gone are the days when knowledge sheet of just sql SQL Injection or XSS could help you land a lucrative high- paying InfoSec job.There is many sheet differnet variations you would login have probably have to try to make this exploit work ( sql especially if it is sql a blind SQL exploit).SQL injection usually occurs when you ask a user for input, like their.ゲストブック/ コメントの例.Submit Text Post.Get an ad- free experience with special benefits, and directly support Reddit.get reddit premium.SQL Injection Cheat.Many web applications have an authentication system: a user provides a user name and password, the web application checks them and stores the corresponding user id in the session hash.Login # 1 Login # 2 Login # 3 Login # 4."
                }
              ]
            }
          ]
        }
      }
    ]
  }
]
}
}

この例では、引用ごとにサポートスコア（groundingScore）が返されます。

根拠のある回答のみを表示する

次のコマンドは、コーパス（データストア内の情報）において根拠があると見なされる回答のみを返す方法を示しています。グラウンディングが不十分な回答は除外されます。

グラウンディングサポートスコアに対して低レベルまたは高レベルのしきい値を選択します。回答がそのレベルを満たしているか、それを超えている場合にのみ、回答が返されます。2 つのフィルタしきい値付きと、しきい値なしを試して、ユーザーにとって最適な結果をもたらすフィルタレベルを判断できます。

REST

サポートスコアのしきい値を満たす場合にのみ回答を返すには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "groundingSpec": {
           "filteringLevel": "FILTER_LEVEL"
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
FILTER_LEVEL: グラウンディングサポートスコアに基づいて回答をフィルタリングするための列挙型。FILTERING_LEVEL_LOW と FILTERING_LEVEL_HIGH を指定できます。filteringLevel が含まれていない場合、回答にサポートスコアフィルタは適用されません。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{
    "query": { "text": "When can an NCD be made?"},
    "groundingSpec": {
       "filtering_level": "FILTERING_LEVEL_HIGH"
    }
  }'
  
  {
    answer {
      state: SUCCEEDED
      answer_text: "We do not have a summary for your query."
      steps {
        state: SUCCEEDED
        description: "Rephrase the query and search."
        actions {
          search_action {
            query: "test?"
          }
          observation {
            search_results {
              document: "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7f5cfde02"
              uri: "gs://my-bucket-123/data/CoverageDocumentation.pdf"
              title: "ABC345_0101"
              chunk_info {
                content: "This notice implements part of section 731 of the Medicare Prescription Drug, Improvement, and Modernization Act of 2003 by describing a method of developing, and making available to the public, guidance documents under the Medicare program… "
              }
    ...
            search_results {
              document: "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7f5cfde02"
              uri: "gs://my-bucket-123/data/CoverageDocumentation.pdf"
              title: "ABC345_0101"
              chunk_info {
                content: "For the purposes of this notice, the term guidance documents means documents prepared for our staff, potential requestors of National Coverage Determinations, and other interested parties explaining the NCD process… "
              }
            }
          }
        }
      }
      answer_skipped_reasons: LOW_GROUNDED_CONTENT
    }

この例では、上限しきい値が満たされていないため、回答は返されません。

回答モデルを指定する

次のコマンドは、回答の生成に使用されるモデルバージョンを変更する方法を示しています。

サポートされているモデルについては、回答生成モデルのバージョンとライフサイクルをご覧ください。

REST

デフォルトモデルとは異なるモデルを使用して回答を生成する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
           "modelSpec": {
              "modelVersion": "MODEL_VERSION",
           }
         }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
MODEL_VERSION: 回答の生成に使用するモデルのバージョン。詳細については、回答生成モデルのバージョンとライフサイクルをご覧ください。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{"query": { "text": "Compare bigquery with spanner database?"},   "answerGenerationSpec": {
     "modelSpec": {
     "modelVersion": "preview",
     }
   } }'
   
   {
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "Cloud Spanner is a fully managed relational database optimized for transactional workloads. BigQuery is a serverless, highly scalable, and cost-effective multi-cloud data warehouse designed for business agility. BigQuery is optimized for ad-hoc analysis and reporting. Both Spanner and BigQuery are built on Google's distributed storage system, Colossus, and their internal cluster management system, Borg. They are also built on Jupiter, Google's in-house custom network hardware and software.\n\nBigQuery can query data stored in Spanner in real time without moving or copying the data. This is possible with BigQuery's query federation support. To run a federated query, you need to configure an external data source in BigQuery that points to the intended Spanner instance. You can then write queries that can be used to populate a BigQuery table on demand or scheduled to run as needed. You can also join the query with another BigQuery result set dynamically.\n\nYou can also use Dataflow to copy data from Spanner to BigQuery. Dataflow is a service that can be used to ingest Spanner data into BigQuery. This is useful for more complex transformations or external dependencies. For example, an online gaming company might use Spanner to store game data and BigQuery to perform analytics on player behavior. They can replicate data from Spanner into BigQuery and perform analytics against local data, or they can use federated queries to retrieve data from Spanner on-demand.\n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Compare bigquery with spanner database?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/ecc0e7547253f4ca3ff3328ce89995af",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/how-spanner-and-bigquery-work-together-handle-transactional-and-analytical-workloads",
                  "title": "How Spanner and BigQuery work together to handle transactional and analytical workloads | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "Using Cloud \u003cb\u003eSpanner\u003c/b\u003e and \u003cb\u003eBigQuery\u003c/b\u003e also allows customers to build their \u003cb\u003edata\u003c/b\u003e clouds using Google Cloud, a unified, open approach to \u003cb\u003edata\u003c/b\u003e-driven transformation ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d7e238f73608a860e00b752ef80e2941",
                  "uri": "https://cloud.google.com/blog/products/databases/cloud-spanner-gets-stronger-with-bigquery-federated-queries",
                  "title": "Cloud Spanner gets stronger with BigQuery-federated queries | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "As enterprises compete for market share, their need for real-time insights has given rise to increased demand for transactional \u003cb\u003edatabases\u003c/b\u003e to support \u003cb\u003edata\u003c/b\u003e ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/e10a5a3c267dc61579e7c00fefe656eb",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/replicating-cloud-spanner-bigquery-scale",
                  "title": "Replicating from Cloud Spanner to BigQuery at scale | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "... \u003cb\u003eSpanner data\u003c/b\u003e into \u003cb\u003eBigQuery\u003c/b\u003e for analytics. In this post, you will learn how to efficiently use this feature to replicate large tables with high throughput ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/648c220055c1d2ac369165007d9f6650",
                  "uri": "https://cloud.google.com/blog/products/databases/choosing-cloud-spanner-for-game-development",
                  "title": "Choosing Cloud Spanner for game development | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "To get started with \u003cb\u003eSpanner\u003c/b\u003e, create a \u003cb\u003edatabase\u003c/b\u003e, or try it out with a \u003cb\u003eSpanner\u003c/b\u003e Qwiklab. ... AlloyDB \u003cb\u003evs\u003c/b\u003e. ... SQL for SQL Server to \u003cb\u003eBigQuery\u003c/b\u003e. By Alexander ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "answerQueryToken": "NMwKDAivmvy2BhCxnsqdARIkNjZkOTQ0NWEtMDAwMC0yMTBkLTllNmItZjQwMzA0NWRiZDMw"
}

この例では、デフォルトモデルではなく、モデルのプレビュー版を使用しています。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

カスタムプリアンブルを指定する

次のコマンドは、生成された回答のプリアンブルの設定方法を示しています。プリアンブルには、回答をカスタマイズするための自然言語の指示が含まれています。長さ、詳細レベル、出力のスタイル（「シンプル」など）、出力の言語、回答の焦点、形式（表、箇条書き、XML など）などのカスタマイズをリクエストできます。たとえば、「10 歳の子どもに説明するように説明して」のようなプリアンブルを使用できます。

プリアンブルは、生成される回答の品質に大きな影響を与える可能性があります。プリアンブルに記述する内容と適切なプリアンブルの例については、カスタムプリアンブルについてをご覧ください。

REST

デフォルトモデルとは異なるモデルを使用して回答を生成する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
           "promptSpec": {
               "preamble": "PREAMBLE",
           }
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
PREAMBLE: 回答をカスタマイズするための自然言語の指示。たとえば、show the answer format in an ordered list や give a very detailed answer を試します。

コマンドの例と結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)"
    -H "Content-Type: application/json"
    "https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
    -d '{
          "query": { "text": "what is bigquery?"},
          "answerGenerationSpec": {
             "promptSpec": {
                "preamble": "Explain like you are a ten years old
                kid",
     }
  }
}'

{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "BigQuery is like a super-powered storage space for your data, but it's in the cloud, not on your computer. It's like a giant warehouse for all your information, but you don't have to build or manage it yourself. You can use BigQuery to find patterns and insights in your data, like figuring out what people like to buy or how many people visited your website. It's like having a super-smart assistant that can help you understand your data better. BigQuery is really good at working with lots of data, even billions of pieces of information. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "What is BigQuery?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/2d032dc582689e8c0ecea7fc7bfa3189",
                  "uri": "https://cloud.google.com/bigquery",
                  "title": "BigQuery enterprise data warehouse | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eBigQuery\u003c/b\u003e is a fully managed, AI-ready data analytics platform that helps you maximize value from your data and is designed to be multi-engine, multi-format, ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/4474f4a5a18ecd611dedfe323dfe55d9",
                  "uri": "https://cloud.google.com/bigquery/docs/introduction",
                  "title": "BigQuery overview | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eBigQuery\u003c/b\u003e is a fully managed, AI-ready data platform that helps you manage and analyze your data with built-in features like machine learning, search, ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/c840fdef90d86328f13bbedbdbf0ac10",
                  "uri": "https://cloud.google.com/bigquery/docs/query-overview",
                  "title": "Overview of BigQuery analytics | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eBigQuery\u003c/b\u003e lets you save queries and share queries with others. When you save a query, it can be private (visible only to you), shared at the project level ( ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/5cbfab8ce1d5f6ffe45fef22900c9a00",
                  "uri": "https://cloud.google.com/curated-resources/bigquery",
                  "title": "Guides BigQuery | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eBigQuery\u003c/b\u003e is Google Cloud's fully managed and serverless enterprise data warehouse solution, designed to help you make informed decisions quickly, so you can ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "answerQueryToken": "NMwKDAi8hN-2BhC0jMCPARIkNjZkN2I4MzItMDAwMC0yMTliLTkxN2EtMDg5ZTA4MjA0YjFj"
}

この例では、プリアンブルはデフォルトよりもシンプルな回答をリクエストします。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

answer メソッドでは、関連する質問を提示することもできます。ユーザーは、独自の質問を入力する代わりに、提示された質問を選択できます。たとえば、「メキシコで休暇を過ごすのに最適な時期はいつですか？」と質問すると、質問への回答に加えて、「メキシコで休暇を過ごすのに最も安い月はいつですか？」や「メキシコの観光シーズンはいつですか？」など、ユーザーが尋ねる可能性のある他の質問を提案できます。

関連する質問を受け取るには、各クエリで関連する質問を含めるように指定する必要があります。関連する質問は、レスポンスで文字列の配列として返されます。

回答がスキップされ、レスポンスで answerSkippedReasons フィールドが返された場合、関連する質問は出力に含まれません。

始める前に

アプリの [生成された回答] オプションがオンになっていることを確認します。

手順

次のコマンドは、回答に関連する質問を含めるようリクエストする方法を示しています。

REST

生成された回答に関連する質問を取得するには:

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "relatedQuestionsSpec": { "enable": true }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。たとえば、「エージェント検索にインポートできるデータの種類は？」などです。

コマンドの例と部分的な結果

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "what kinds of data can I import?"},
        "session": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10291252835232308789",
        "searchSpec":{ "searchParams": {"filter": ""} },
        "relatedQuestionsSpec": { "enable": true }
}'

{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "You can import various kinds of data, depending on the type of data store you create. Each data store can contain one type of data. The types of data stores include website data, structured data, structured data for media, unstructured data, and healthcare data. ... You can import FHIR R4 data into a data store in the following ways: Batch import (a one-time import).\n",
    "relatedQuestions": [
      "Can you provide examples of structured data formats?",
      "What file types are supported for unstructured data?",
      "How does website data ingestion work?",
      "What are the limitations of healthcare data import?",
      "Is there a size limit for importing data?"
    ],
    ...
  }
}

この例では、relatedQuestionsSpec オブジェクトの enabled フィールドが true に設定され、レスポンスには relatedQuestions 配列に関連する質問が含まれています。

引用を含める

次のコマンドは、回答に引用を含めるようリクエストする方法を示しています。

REST

デフォルトモデルとは異なるモデルを使用して回答を生成する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
           "includeCitations": INCLUDE_CITATIONS
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
INCLUDE_CITATIONS: 回答に引用メタデータを含めるかどうかを指定します。デフォルト値は false です。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{"query": { "text": "What is SQL"}, "answerGenerationSpec": {
"includeCitations": true
}
}'

{
"answer": {
"state": "SUCCEEDED",
"answerText": "SQL stands for Structured Query Language and is a programming language used to manage, query, and retrieve data in a relational database. It is the standard language used by relational database management systems (RDBMS) such as PostgreSQL, SQL Server, MySQL, and Oracle Database. SQL statements are written in a statement format for queries and other database operations, allowing users to manipulate data in relational database tables. SQL is used to create and update the structure of tables, read and write data, manage user permissions, and perform administrative tasks. While originally created for relational databases, SQL is now a foundation for many technology capabilities, making SQL knowledge essential for many technology roles. \n",
"citations": [
{
"endIndex": "137",
"sources": [
{
"referenceId": "0"
},
{
"referenceId": "1"
}
]
},
{
"startIndex": "138",
"endIndex": "437",
"sources": [
{
"referenceId": "3"
}
]
},
{
"startIndex": "438",
"endIndex": "575",
"sources": [
{
"referenceId": "2"
}
]
},
{
"startIndex": "576",
"endIndex": "742",
"sources": [
{
"referenceId": "3"
}
]
}
],
"references": [
{
"chunkInfo": {
"content": "There may be a second table that stores visit information. A relational database uses a unique ID for each row to maintain the linked patient information across the two tables. This way, you can quickly look up the visits of each patient. Sign up for a free trial for any of Google Cloud's SQL databases, which include AlloyDB, Cloud SQL, and Spanner. Get started for free What is SQL? SQL (Structured Query Language) is a programming language used to store, retrieve, and manage data in a relational database. SQL statements are English-like, making the language accessible to software developers, data analysts, and other practitioners. Benefits of SQL databases Enterprises choose SQL databases for being: Efficient. Relational databases are incredibly efficient in managing complex queries. Fast. SQL databases can retrieve large amounts of data, quickly. This makes them highly desirable for real-time transactional data. Reliable. SQL databases provide a high degree of data integrity and are ACID-compliant. SQL database engines There are numerous SQL database engines (products) used to build software applications. Some of the most popular include PostgreSQL, MySQL, SQL Server, and Oracle. Some database engines are open source while others are commercial offerings. ",
"relevanceScore": 0.9,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/7218ff4f57328d86059246d4af3a9953",
"uri": "https://cloud.google.com/discover/what-are-sql-databases",
"title": "SQL Databases | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "PostgreSQL vs. SQL Server: What's the difference? | Google Cloud Page Contents Topics PostgreSQL vs. SQL PostgreSQL vs SQL Server: What are the key differences? Trying to find the right database for your applications? When it comes to choosing a database technology, the most common SQL options to consider are PostgreSQL vs. SQL Server. While both systems share many core features, there are some key differences—the major one being that PostgreSQL is open source and SQL Server is owned by Microsoft. Today, it is more vital than ever for companies to be able to manage, store, and activate data for modern business operations. With the growing assortment of databases available to choose from, it can be overwhelming to pick the right one for your applications. The most important thing to remember is that no single database will be a good match for every project requirement, so it's critical to understand the option that will work best for your specific use case. So, what is PostgreSQL vs. SQL Server? In this short guide, we'll discuss the basic differences between PostgreSQL and SQL Server. Get started for freeStay informed What is SQL? Structured Query Language or SQL, as it's more commonly known, is a programming language used to manage, query, and retrieve data in a relational database. ",
"relevanceScore": 0.8,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7cd9afab1282a9f57cdcee1885bb4c6",
"uri": "https://cloud.google.com/learn/postgresql-vs-sql",
"title": "PostgreSQL vs. SQL Server: What's the difference? | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "SQL Databases | Google Cloud Page Contents Topics What are SQL databases? What are SQL databases? A SQL database, also known as a relational database, is a system that stores and organizes data into highly structured tables of rows and columns. These databases offer Structured Query Language (SQL) to read and write the data, and are categorized as relational database management systems (RDBMS). SQL statements are used to create and update the structure of tables, read and write data, manage user permissions, and perform administrative tasks. For example, a CREATE statement is used to create a table, an INSERT statement adds a new row to a table, and a SELECT statement performs a database query. Statements that make structural or administrative changes are usually reserved for software developers and administrators, while read and write operations are performed by end-user applications. A relational database maintains the ability to link information across multiple tables. This format makes it easy to quickly gain insights about the relationships between various columns or data points in these tables. A relational database can create indexes for particular columns for faster lookups. For example, a healthcare facility might maintain a table containing rows of patient information, where each row is one patient and the columns contain data points, such as the patient's name, insurance information, and contact details. ",
"relevanceScore": 0.8,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/7218ff4f57328d86059246d4af3a9953",
"uri": "https://cloud.google.com/discover/what-are-sql-databases",
"title": "SQL Databases | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "It is the standard language used by relational database management systems (RDBMS), including PostgreSQL, SQL Server, MySQL, and Oracle Database. SQL typically uses commands written in statement format for queries and other database operations, which allow users to manipulate data in relational database tables. While originally created for relational databases, SQL acts as a foundation for many of today's technology capabilities, making SQL knowledge an essential skill for many technology roles today, including data analysts, database engineers, and even backend programming. However, you will find that there are different variants of SQL depending on the database or database management system that you choose. What is Microsoft SQL Server? SQL Server is a leading RDBMS that is built on top of SQL and developed by Microsoft. It is used to manage and store data to support numerous enterprise use cases for business intelligence, transaction processing, data analytics, and machine learning services. SQL Server has a row-based table structure that allows you to connect related data elements from different tables without having to store data multiple times in a database. In general, Microsoft SQL Server is known for its high availability, fast performance when handling large workloads, and easy integration with other applications to gain business intelligence across your entire data estate. For more information, we recommend reviewing the official SQL Server documentation. ",
"relevanceScore": 0.8,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7cd9afab1282a9f57cdcee1885bb4c6",
"uri": "https://cloud.google.com/learn/postgresql-vs-sql",
"title": "PostgreSQL vs. SQL Server: What's the difference? | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "Send feedback The GoogleSQL language in Spanner bookmark_borderbookmark Stay organized with collections Save and categorize content based on your preferences. Dismiss Got it GoogleSQL is the new name for Google Standard SQL! New name, same great SQL dialect. This page provides an overview of supported statements in GoogleSQL. GoogleSQL is an ANSI compliant Structured Query Language (SQL) which includes the following types of supported statements: Query statements, also known as Data Query Language (DQL) statements, are the primary method to analyze data in Spanner. They scan one or more tables or expressions and return the computed result rows. Data Definition Language (DDL) statements let you create and modify database objects such as tables, views, and database roles. Data Manipulation Language (DML) statements enable you to update, insert, and delete data from your Spanner tables. Data Access Control statements let you grant and revoke access privileges at the table and column level. Transaction Control statements allow you to manage transactions for data modifications. Was this helpful? Send feedback Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates. ",
"relevanceScore": 0.7,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/0c5c094170756eeb6bdfec6eb5c7d081",
"uri": "https://cloud.google.com/spanner/docs/reference/standard-sql/overview",
"title": "The GoogleSQL language in Spanner | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "FAQ Expand all What is Cloud SQL? Cloud SQL is a service that delivers fully managed relational databases in the cloud. It offers MySQL, PostgreSQL, and SQL Server database engines. How is Cloud SQL different from other cloud databases? Cloud SQL is valued for its openness, ease of use, security, cost-efficiency, and Google Cloud integration—in fact, more than 95% of Google Cloud's top 100 customers use it. If you're comparing PostgreSQL options on Google Cloud, view our comparison chart. What's the difference between the Enterprise and Enterprise Plus editions? For PostgreSQL, the Enterprise Plus edition brings enhanced availability, performance, and data protection capabilities. Specifically, it provides a 99.99% availability SLA with near-zero downtime maintenance, optimized hardware and software configurations, intelligent data caching for read-intensive transactional workloads, a configurable data cache option and 35 days of log retention. For MySQL, the Enterprise Plus edition brings enhanced availability, performance, and data protection capabilities. Specifically, it provides a 99.99% availability SLA with near-zero downtime maintenance, optimized hardware and software configurations, intelligent data caching for read-intensive transactional workloads, a configurable data cache option, 35 days of log retention and advanced disaster recovery capabilities like orchestrated failover and switchback. ",
"relevanceScore": 0.7,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/931f2c8e19ed54a407857f1cad3b5aaa",
"uri": "https://cloud.google.com/sql",
"title": "Cloud SQL for MySQL, PostgreSQL, and SQL Server | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "PostgreSQL versus SQL PostgreSQL is an open-source, object-relational database (ORDBMS) designed for enterprise-level performance and is valued for its reliability and robust features. Its long history of development and its use of SQL makes it one of the most popular open source databases worldwide. Its default procedural language is an extension of pgSQL (PL/pgSQL), with procedural language extensions of Tcl, Perl, and Python included in the standard distribution (written as PL/Tcl, PL/Perl, and PL/Python). Many more languages are supported through extensions, including Java, Ruby, C, C++, Delphi, and JavaScript. For a more in-depth comparison, visit our PostgreSQL versus SQL guide. MySQL versus SQL MySQL is a popular open source relational database created in 1995 and currently sponsored by Oracle. It supports SQL queries and can be administered either through a graphical user interface (GUI) or a command line. MySQL can be deployed manually on a physical machine or through a cloud service provider. Enterprises are increasingly choosing fully managed services to reduce the maintenance burden of their databases. What is SQL Server? SQL Server is a Microsoft-owned database that runs SQL queries. Dive into the differences between PostgreSQL and SQL Server. ",
"relevanceScore": 0.6,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/7218ff4f57328d86059246d4af3a9953",
"uri": "https://cloud.google.com/discover/what-are-sql-databases",
"title": "SQL Databases | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "Send feedback On this page BigQuery SQL dialects Changing from the default dialect What's next Introduction to SQL in BigQuery bookmark_borderbookmark Stay organized with collections Save and categorize content based on your preferences. Dismiss Got it GoogleSQL is the new name for Google Standard SQL! New name, same great SQL dialect. This document provides an overview of supported statements and SQL dialects in BigQuery. GoogleSQL is an ANSI compliant Structured Query Language (SQL) which includes the following types of supported statements: Query statements, also known as Data Query Language (DQL) statements, are the primary method to analyze data in BigQuery. They scan one or more tables or expressions and return the computed result rows. Procedural language statements are procedural extensions to GoogleSQL that allow you to execute multiple SQL statements in one request. Procedural statements can use variables and control-flow statements, and can have side effects. Data Definition Language (DDL) statements let you create and modify database objects such as tables, views, functions, and row-level access policies. Data Manipulation Language (DML) statements enable you to update, insert, and delete data from your BigQuery tables. Data Control Language (DCL) statements let you control BigQuery system resources such as access and capacity. Transaction Control Language (TCL) statements allow you to manage transactions for data modifications. ",
"relevanceScore": 0.6,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/2f6fc3e29873518196cb50195d7ded45",
"uri": "https://cloud.google.com/bigquery/docs/introduction-sql",
"title": "Introduction to SQL in BigQuery | Google Cloud"
}
}
},
{
"chunkInfo": {
"content": "Database administration Cloud SQL pricing Connect to a Cloud SQL managed database Cloud SQL updates Configuration updates System updates What's next Home Cloud SQL Documentation Guides Was this helpful? Send feedback Cloud SQL overview bookmark_borderbookmark Stay organized with collections Save and categorize content based on your preferences. Dismiss Got it On this page Database configurations with Cloud SQL Use cases for Cloud SQL What Cloud SQL provides What is a Cloud SQL instance? Database administration Cloud SQL pricing Connect to a Cloud SQL managed database Cloud SQL updates Configuration updates System updates What's next Cloud SQL is a fully managed relational database service for MySQL, PostgreSQL, and SQL Server. This frees you from database administration tasks so that you have more time to manage your data. This page discusses basic concepts and terminology for Cloud SQL, which provides SQL data storage for Google Cloud. For a more in-depth explanation of key concepts, see the key terms and features pages. For information about how Cloud SQL databases compare with one another, see Cloud SQL feature support by database engine. Database configurations with Cloud SQL The following video shows you the benefits of using Cloud SQL. ",
"relevanceScore": 0.6,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/4098ae11bfa400e8f1b8e9ba59d2b71b",
"uri": "https://cloud.google.com/sql/docs/introduction",
"title": "Cloud SQL overview"
}
}
},
{
"chunkInfo": {
"content": "Cloud SQL documentation View all product documentation Cloud SQL is a fully-managed database service that helps you set up, maintain, manage, and administer your relational databases on Google Cloud Platform. You can use Cloud SQL with MySQL, PostgreSQL, or SQL Server. Not sure what database option is right for you? Learn more about our database services. Learn more about Cloud SQL. Documentation resources Find quickstarts and guides, review key references, and get help with common issues. format_list_numbered Guides Cloud SQL overview Database engine feature support MySQL PostgreSQL SQL Server find_in_page Reference gcloud commands REST API Client libraries info Resources Pricing Release notes Resources Try Cloud SQL for yourself Create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads. Try Cloud SQL free Was this helpful? Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates. Last updated 2024-08-29 UTC. ",
"relevanceScore": 0.5,
"documentMetadata": {
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/37935181d99a6ad3b4897e673a7a7986",
"uri": "https://cloud.google.com/sql/docs",
"title": "Cloud SQL documentation"
}
}
}
],
"steps": [
{
"state": "SUCCEEDED",
"description": "Rephrase the query and search.",
"actions": [
{
"searchAction": {
"query": "What is SQL?"
},
"observation": {
"searchResults": [
{
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/7218ff4f57328d86059246d4af3a9953",
"uri": "https://cloud.google.com/discover/what-are-sql-databases",
"title": "SQL Databases | Google Cloud",
"snippetInfo": [
{
"snippet": "\u003cb\u003eSQL\u003c/b\u003e (Structured Query Language) is a programming language used to store, retrieve, and manage data in a relational database. \u003cb\u003eSQL\u003c/b\u003e statements are English-like, ...",
"snippetStatus": "SUCCESS"
}
]
},
{
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7cd9afab1282a9f57cdcee1885bb4c6",
"uri": "https://cloud.google.com/learn/postgresql-vs-sql",
"title": "PostgreSQL vs. SQL Server: What's the difference? | Google Cloud",
"snippetInfo": [
{
"snippet": "\u003cb\u003eSQL\u003c/b\u003e typically uses commands written in statement format for queries and other database operations, which allow users to manipulate data in relational database ...",
"snippetStatus": "SUCCESS"
}
]
},
{
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/3afdede140d0906c2146a2f2b3a7821e",
"uri": "https://cloud.google.com/blog/topics/developers-practitioners/what-cloud-sql",
"title": "What is Cloud SQL? | Google Cloud Blog",
"snippetInfo": [
{
"snippet": "It is a fully managed relational database for MySQL, PostgreSQL and \u003cb\u003eSQL\u003c/b\u003e Server. It reduces maintenance cost and automates database provisioning, storage ...",
"snippetStatus": "SUCCESS"
}
]
},
...
{
"document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/0c5c094170756eeb6bdfec6eb5c7d081",
"uri": "https://cloud.google.com/spanner/docs/reference/standard-sql/overview",
"title": "The GoogleSQL language in Spanner | Google Cloud",
"snippetInfo": [
{
"snippet": "\u003cb\u003eGoogleSQL\u003c/b\u003e is the new name for Google Standard \u003cb\u003eSQL\u003c/b\u003e! New name, same great \u003cb\u003eSQL\u003c/b\u003e dialect. This page provides an overview of supported statements in \u003cb\u003eGoogleSQL\u003c/b\u003e.",
"snippetStatus": "SUCCESS"
}
]
}
]
}
}
]
}
]
},
"answerQueryToken": "NMwKDAiFm_y2BhC_nfrYAxIkNjZkYjg3NjItMDAwMC0yZTBkLTg0ZDAtMDg5ZTA4MmRjYjg0"
}

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

回答の言語コードを設定する

次のコマンドは、回答の言語コードを設定する方法を示しています。

REST

デフォルトモデルとは異なるモデルを使用して回答を生成する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
           "answerLanguageCode": "ANSWER_LANGUAGE_CODE"
           }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
ANSWER_LANGUAGE_CODE: 回答の言語コード。BCP47: 言語を識別するためのタグで定義されている言語タグを使用します。

コマンドの例と結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{"query": { "text": "What is SQL"},   "answerGenerationSpec": {
  "answerLanguageCode": "es"
     }
}'

{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "SQL, que significa Structured Query Language, es un lenguaje de programación utilizado para almacenar, recuperar y administrar datos en una base de datos relacional. Las instrucciones de SQL son similares al inglés, lo que hace que el lenguaje sea accesible para desarrolladores de software, analistas de datos y otros profesionales. Las bases de datos SQL se utilizan para administrar y almacenar datos para apoyar numerosos casos de uso empresariales, como la inteligencia empresarial, el procesamiento de transacciones, el análisis de datos y los servicios de aprendizaje automático. SQL es el lenguaje estándar utilizado por los sistemas de gestión de bases de datos relacionales (RDBMS), incluidos PostgreSQL, SQL Server, MySQL y Oracle Database. SQL se utiliza para crear y actualizar la estructura de las tablas, leer y escribir datos, administrar los permisos de los usuarios y realizar tareas administrativas. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "What is SQL?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/7218ff4f57328d86059246d4af3a9953",
                  "uri": "https://cloud.google.com/discover/what-are-sql-databases",
                  "title": "SQL Databases | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eSQL\u003c/b\u003e (Structured Query Language) is a programming language used to store, retrieve, and manage data in a relational database. \u003cb\u003eSQL\u003c/b\u003e statements are English-like, ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7cd9afab1282a9f57cdcee1885bb4c6",
                  "uri": "https://cloud.google.com/learn/postgresql-vs-sql",
                  "title": "PostgreSQL vs. SQL Server: What's the difference? | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eSQL\u003c/b\u003e typically uses commands written in statement format for queries and other database operations, which allow users to manipulate data in relational database ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/3afdede140d0906c2146a2f2b3a7821e",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/what-cloud-sql",
                  "title": "What is Cloud SQL? | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "It is a fully managed relational database for MySQL, PostgreSQL and \u003cb\u003eSQL\u003c/b\u003e Server. It reduces maintenance cost and automates database provisioning, storage ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/0c5c094170756eeb6bdfec6eb5c7d081",
                  "uri": "https://cloud.google.com/spanner/docs/reference/standard-sql/overview",
                  "title": "The GoogleSQL language in Spanner | Google Cloud",
                  "snippetInfo": [
                    {
                      "snippet": "\u003cb\u003eGoogleSQL\u003c/b\u003e is the new name for Google Standard \u003cb\u003eSQL\u003c/b\u003e! New name, same great \u003cb\u003eSQL\u003c/b\u003e dialect. This page provides an overview of supported statements in \u003cb\u003eGoogleSQL\u003c/b\u003e.",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "answerQueryToken": "NMwKDAjim_y2BhDftIjEAhIkNjZkOTQ0NWQtMDAwMC0yMTBkLTllNmItZjQwMzA0NWRiZDMw"
}

この例では、ソースドキュメントは英語ですが、回答はスペイン語で提供されています。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.api_core.client_options import ClientOptions
from google.cloud import discoveryengine_v1 as discoveryengine

# TODO(developer): Uncomment these variables before running the sample.
# project_id = "YOUR_PROJECT_ID"
# location = "YOUR_LOCATION"                    # Values: "global", "us", "eu"
# engine_id = "YOUR_APP_ID"


def answer_query_sample(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.AnswerQueryResponse:
    #  For more information, refer to:
    # https://cloud.google.com/generative-ai-app-builder/docs/locations#specify_a_multi-region_for_your_data_store
    client_options = (
        ClientOptions(api_endpoint=f"{location}-discoveryengine.googleapis.com")
        if location != "global"
        else None
    )

    # Create a client
    client = discoveryengine.ConversationalSearchServiceClient(
        client_options=client_options
    )

    # The full resource name of the Search serving config
    serving_config = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/servingConfigs/default_serving_config"

    # Optional: Options for query phase
    # The `query_understanding_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/QueryUnderstandingSpec
    query_understanding_spec = discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec(
        query_rephraser_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryRephraserSpec(
            disable=False,  # Optional: Disable query rephraser
            max_rephrase_steps=1,  # Optional: Number of rephrase steps
        ),
        # Optional: Classify query types
        query_classification_spec=discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec(
            types=[
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.ADVERSARIAL_QUERY,
                discoveryengine.AnswerQueryRequest.QueryUnderstandingSpec.QueryClassificationSpec.Type.NON_ANSWER_SEEKING_QUERY,
            ]  # Options: ADVERSARIAL_QUERY, NON_ANSWER_SEEKING_QUERY or both
        ),
    )

    # Optional: Options for answer phase
    # The `answer_generation_spec` below includes all available query phase options.
    # For more details, refer to https://cloud.google.com/generative-ai-app-builder/docs/reference/rest/v1/AnswerGenerationSpec
    answer_generation_spec = discoveryengine.AnswerQueryRequest.AnswerGenerationSpec(
        ignore_adversarial_query=False,  # Optional: Ignore adversarial query
        ignore_non_answer_seeking_query=False,  # Optional: Ignore non-answer seeking query
        ignore_low_relevant_content=False,  # Optional: Return fallback answer when content is not relevant
        model_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.ModelSpec(
        # Use the 2026 stable production model for answer generation
        model_version="gemini-2.5-flash/answer_gen/stable",
        ),
        prompt_spec=discoveryengine.AnswerQueryRequest.AnswerGenerationSpec.PromptSpec(
            preamble="Give a detailed answer.",  # Optional: Natural language instructions for customizing the answer.
        ),
        include_citations=True,  # Optional: Include citations in the response
        answer_language_code="en",  # Optional: Language code of the answer
    )

    # Initialize request argument(s)
    request = discoveryengine.AnswerQueryRequest(
        serving_config=serving_config,
        query=discoveryengine.Query(text="What is Vertex AI Search?"),
        session=None,  # Optional: include previous session ID to continue a conversation
        query_understanding_spec=query_understanding_spec,
        answer_generation_spec=answer_generation_spec,
        user_pseudo_id="user-pseudo-id",  # Optional: Add user pseudo-identifier for queries.
    )

    # Make the request
    response = client.answer_query(request)

    # Handle the response
    print(response)

    return response

回答をパーソナライズする

ユーザーに関する特定の情報（プロフィール内のデータなど）がある場合は、endUserMetadata オブジェクトでその情報を指定して、クエリ結果をユーザーに合わせてカスタマイズできます。

たとえば、ログインしているユーザーが携帯電話のアップグレードに関する情報を検索している場合、現在の携帯電話のモデルや携帯通信プランなどのプロフィール情報に基づいて、生成された回答をパーソナライズできます。

クエリを実行するユーザーの個人情報を追加し、その個人情報を考慮した回答を生成するには、次の操作を行います。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
-d '{
    "query": { "text": "QUERY"},
    "endUserSpec": {
       "endUserMetadata": [
         {
           "chunkInfo": {
              "content": "PERSONALIZED_INFO",
              "documentMetadata":  { "title": "INFO_DESCRIPTION"}
           }
         }
       ]
    }
  }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
PERSONALIZATION_INFO: クエリを実行しているユーザーに固有の情報を含む文字列。たとえば、This customer has a Pixel 6 Pro purchased over a period of 24-months starting 2023-01-15. This customer is on the Business Plus International plan. No payment is due at this time. この文字列の長さの上限は 8,000 文字です。
INFO_DESCRIPTION: パーソナライズ情報を簡単に説明する文字列（例: Customer profile data, including model, plan, and billing status.）。モデルは、クエリに対するカスタマイズされた回答を生成する際に、この説明とパーソナライズ情報の両方を使用します。

コマンドの例と結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "Can I upgrade my phone now?"},
        "answerGenerationSpec": { "includeCitations": true }
        "endUserSpec": {
           "endUserMetadata": [
             {
               "chunkInfo": {
                 "content": "This customer has a Pixel 6 Pro purchased over a period of 24-months starting 2023-01-15. This customer is on the Business Plus International plan. No payment is due at this time.",
                 "documentMetadata":  { "title": "Customer profile data, including model, plan, and billing status."}
               }
             }
           ]
        }
      }'

{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "Yes, you qualify for the yearly device upgrade because you've completed your 24-month payment plan.[1,3]  Since your account is fully paid you will not need to pay any additional monthly fees…  \n",
    "citations": [
      {
        "end_index": 99
        "sources": [
           {
              "reference_id": "0"
           }
        ]
      },
      {
        "start_index": "100"
        "end_index": "240"
        "sources": [
           {
              "reference_id": "0"
           }
        ]
      },
      ...
    ]
    "references": [
      {
        "chunk_info": {
          "content":
          "+ "This customer has a Pixel 6 Pro purchased over a period of 24-months starting 2023-01-15. This customer is on the Business Plus International plan. No payment is due at this time.",
          "relevance_score": 0.3
          "documentMetadata":  {
            "title": "Customer profile data, including model, plan, and billing status."
          }
      {
        "chunk_info": {
          "content":
          "+ "For Cymbal Mobile upgrades, you can upgrade when you've paid off at least half of your current device's cost and have had it for 8 months…",
          "relevance_score": 0.8
          "documentMetadata":  {
            "document":
            "projects/123456/locations/us/collections/default_collection/dataStores/my-data-store/branches/0/documents/abcd1234567890"
            "uri": "https://www.example.com/help/device-upgrade"
            "title": "Upgrade eligibility"
          }
         }
      }
      ...

この例では、モデルは endUserMetadata を使用して回答をカスタマイズします。この例では、endUserMetadata の重要性を理解できるように引用が含まれています。これは常に最初に引用される参照です。

回答のグラフを生成する

answer メソッドは、グラフを生成して、クエリの回答の一部として返すことができます。

たとえば、「利用可能なデータを使用して、中小企業の支払いの前年比成長率を年ごとにプロットして」のように、回答にグラフを含めるよう具体的にリクエストできます。十分なデータが存在すると判断された場合は、グラフが返されます。通常、グラフとともに回答テキストが返されます。

また、グラフを作成するのに十分なデータがある場合、クエリでグラフが明示的にリクエストされていなくても、回答メソッドはグラフを返す場合があります。たとえば、「2010 年から 2020 年の 10 年間で、安全な飲料水へのアクセス増加に関連して HDI スコアはどの程度改善しましたか？」などです。

回答ごとに生成されるグラフは 1 つのみです。ただし、グラフが複合グラフで、他の小さなグラフが含まれている場合があります。複合グラフの例:

複合グラフには 4 つの小さなグラフが含まれている

制約事項

クエリは英語で入力する必要があります。

一般的な障害シナリオ

回答とともに画像が返されるとは限りません。十分なデータがない場合は、図を生成できません。

その他の障害シナリオには、コード実行の失敗やタイムアウトなどがあります。どちらかの場合は、クエリを言い換えてもう一度お試しください。

始める前に

生成されたグラフをリクエストするクエリを実行する前に、次の操作を行います。

アプリの [生成された回答] オプションがオンになっていることを確認します。
Gemini 2.0 以降のモデルを使用していることを確認します。モデルの詳細については、回答生成モデルのバージョンとライフサイクルをご覧ください。
テーブルや画像が多数含まれるドキュメントを含む非構造化データストアがある場合は、レイアウトパーサーを有効にします。これは必須ではありませんが、結果の品質が向上します。

手順

REST

次のように answer メソッドを呼び出して、データストア内のデータから生成されたグラフを含む回答を返します。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1beta/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
          "model_spec": {
             "model_version": "MODEL_VERSION"
         },
          "multimodalSpec": {
             "imageSource": "IMAGE_SOURCE"
             }
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含む英語の自由形式の文字列。
MODEL_VERSION: モデルバージョン gemini-2.0-flash-001/answer_gen/v1 以降。詳細については、回答生成モデルのバージョンとライフサイクルをご覧ください。
IMAGE_SOURCE: 生成されたグラフを回答に含めるようリクエストする列挙型（FIGURE_GENERATION_ONLY）、または生成されたグラフかデータストアの既存の画像のいずれかを回答に含めることができる列挙型（ALL_AVAILABLE_SOURCES）。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://discoveryengine.googleapis.com/v1beta/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer" \
-d '{
    "query": { "text": "Plot composition of net cash used in investing activities"},
    "answerGenerationSpec": {
      "model_spec": {
         "model_version": "gemini-2.0-flash-001/answer_gen/v1"
         },
         "multimodalSpec": {
         "image_source": "FIGURE_GENERATION_ONLY"
         }
    }
  }'
  
{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "The composition of net cash used in investing activities for the nine months ended September 30, 2020 and 2019 includes several categories. These categories are purchases of property and equipment, purchases of marketable securities, maturities and sales of marketable securities, purchases of non-marketable investments, maturities and sales of non-marketable investments, acquisitions net of cash acquired and purchases of intangible assets, and other investing activities. For the nine months ended September 30, 2020, the net cash used in investing activities totaled $(25,492). For the nine months ended September 30, 2019, the net cash used in investing activities totaled $(24,788).\n\nHere's a breakdown of the specific cash flows for investing activities (in millions):\n\n*   **Purchases of property and equipment:** $(16,802) in 2020 and $(17,496) in 2019\n*   **Purchases of marketable securities:** $(104,932) in 2020 and $(80,968) in 2019\n*   **Maturities and sales of marketable securities:** $97,751 in 2020 and $74,783 in 2019\n*   **Purchases of non-marketable investments:** $(1,864) in 2020 and $(1,499) in 2019\n*   **Maturities and sales of non-marketable investments:** $598 in 2020 and $297 in 2019\n*   **Acquisitions, net of cash acquired, and purchases of intangible assets:** $(368) in 2020 and $(373) in 2019\n*   **Other investing activities:** $125 in 2020 and $468 in 2019",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Plot composition of net cash used in investing activities"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/b3133a895e0404984959736488992b53",
                  "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/boa-tabular/Testing_20201030-alphabet-cash flow statement.pdf",
                  "title": "Testing_20201030-alphabet-cash flow statement",
                  "snippetInfo": [
                    {
                      "snippet": "... \u003cb\u003eNet cash provided by operating activities\u003c/b\u003e 42,447 | 40,093 | | \u003cb\u003eInvesting activities\u003c/b\u003e | | | (16,802) Purchases of property and equipment | (17,496) ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ],
    "blobAttachments": [
      {
        "data": {
          "mimeType": "image/png",
          "data": "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...PrTn86//Mu/5Jhjjql1NACAAeP/AX0CwI314+qiAAAAAElFTkSuQmCC"
        },
        "attributionType": "GENERATED"
      }
      ]
  },
  "answerQueryToken": "NMwKDAjEjOe-BhD-meX6ARIkNjdkNjBhM2QtMDAwMC0yYzU4LTgxYjctMDg5ZTA4MmNhZDgw"

この例では、クエリでプロットがリクエストされ、グラフを生成するのに十分なデータがあったため、生成されたグラフが回答に含まれています。

グラフの画像は、GENERATED の attributionType を含む blobAttachments 出力で返されます。

データストアから既存の画像を取得する

データストアの画像を回答とともに、引用の参照として返すように選択できます。データストアは、レイアウトパーサーが有効になっている非構造化データストアである必要があります。

返された回答にグラフを表示するには、[生成された回答] オプションをオンにする必要があります。

imageSource が CORPUS_IMAGE_ONLY または ALL_AVAILABLE_SOURCES の場合、answer メソッドはデータストアから画像を適宜取得できます。ただし、この機能をオンにしても、画像が常に返されるわけではありません。

回答ごとに 1 枚の画像（最大）が生成されます。引用には複数の画像を含めることができます。

制限事項

使用しているアプリが非構造化データストアに接続されている必要があります。画像はウェブサイトや構造化データストアから返されません。
クエリは英語で入力する必要があります。
レイアウトパーサーによる画像アノテーションは、データストアに適用する必要があります。レイアウトパーサーについては、ドキュメントを解析してチャンクに分割するをご覧ください。

手順

REST

次のように answer メソッドを呼び出して、データストアの画像を含む回答を返します。

次の curl コマンドを実行します。

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1beta/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "answerGenerationSpec": {
          "model_spec": {
             "model_version": "MODEL_VERSION"
          },
          includeCitations: true,
          "multimodalSpec": {
             "imageSource": "IMAGE_SOURCE"
             }
        }
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: クエリする Agent Search アプリの ID。
QUERY: 質問または検索クエリを含む英語の自由形式の文字列。
MODEL_VERSION: モデルバージョン gemini-2.0-flash-001/answer_gen/v1 以降。詳細については、回答生成モデルのバージョンとライフサイクルをご覧ください。
IMAGE_SOURCE: 回答にデータストアの画像を含めることをリクエストする列挙型（CORPUS_IMAGE_ONLY）、または回答にデータストアの画像または生成されたグラフのいずれかを含めることをリクエストする列挙型（ALL_AVAILABLE_SOURCES）。

コマンドの例と部分的な結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \
 -H "Content-Type: application/json" \
 "https://discoveryengine.googleapis.com/v1beta/projects/my-project-123//locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer"   -d '{
        "query": { "text": "Display the ranking of  generations from highest to lowest based on their spending on holiday items using cards."},
        "answerGenerationSpec": {"includeCitations": true,
        "model_spec": {
             "model_version": "gemini-2.0-flash-001/answer_gen/v1"
             },
             "multimodalSpec": {
      "imageSource": "CORPUS_IMAGE_ONLY"
      }
  },
  "searchSpec": {
    "searchParams": {
      "maxReturnResults": 5
    }
  }
}'

{
  "answer": {
    "state": "SUCCEEDED",
    "answerText": "Holiday spending over November-December was weakest among Gen Xers, followed by Millennials, while Baby Boomer spending was relatively more solid. In November, Baby Boomers saw solid spending growth. Both Millennials and Gen X consumers had lower holiday spending than in the same 30-day period last year. Baby Boomers' holiday spending started to pick up from the start of November and continued to rise throughout the month. Meanwhile, Millennials started to ramp up their holiday spending much later in the month. Gen X consumers consistently outspent other generations from late October through the end of November.\n",
    "citations": [
      {
        "endIndex": "146",
        "sources": [
          {
            "referenceId": "3"
          }
        ]
      },
      {
        "startIndex": "147",
        "endIndex": "199",
        "sources": [
          {
            "referenceId": "2"
          }
        ]
      },
      {
        "startIndex": "200",
        "endIndex": "305",
        "sources": [
          {
            "referenceId": "2"
          }
        ]
      },
      {
        "startIndex": "306",
        "endIndex": "426",
        "sources": [
          {
            "referenceId": "2"
          }
        ]
      },
      {
        "startIndex": "427",
        "endIndex": "516",
        "sources": [
          {
            "referenceId": "2"
          }
        ]
      },
      {
        "startIndex": "517",
        "endIndex": "619",
        "sources": [
          {
            "referenceId": "2"
          }
        ]
      }
    ],
    "references": [
      {
        "chunkInfo": {
          "content": "Bank of America aggregated credit/debit card spending per household includes spending from active US households only. Only.., Consumer and Small Business Mel Roasa Vice President, Digital and MarketingBANK OF AMERICA INSTITUTE09 November 20237",
          "relevanceScore": 0.5,
          "documentMetadata": {
            "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f8d7887862167c5daf6c7a30e1d464e0",
            "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1WkoquhDpqHphSnqIVKX45iers7kvmGjZ.pdf",
            "title": "1WkoquhDpqHphSnqIVKX45iers7kvmGjZ",
            "pageIdentifier": "6"
          }
        }
      },
      {
        "chunkInfo": {
          "content": "Bank of America aggregated credit/debit card spending per household includes spending from active US households only. Only... Digital and Marketing Riley Fillius Vice President, Digital and MarketingBANK OF AMERICA INSTITUTE08 February 2024 7",
          "relevanceScore": 0.4,
          "documentMetadata": {
            "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/c4308d8610bcae3b3d2bb916090ef28b",
            "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1B3MJqELqk_S26aWb-KnLdSWwmpuB32os.pdf",
            "title": "1B3MJqELqk_S26aWb-KnLdSWwmpuB32os",
            "pageIdentifier": "6"
          }
        }
      },
...
      {
        "chunkInfo": {
          "content": "Bank of America aggregated credit/debit card spending per household includes spending from active US households only. Only consumer card holders making a minimum of five transactions a month are included in the dataset. Spending from corporate cards are excluded. Data regarding merchants who receive payments are identified and classified by the Merchant Categorization Code (MCC) defined by financial services companies. The data are mapped using proprietary methods from the MCCs to the North American Industry Classification System (NAICS), which is also used by the Census Bureau, in order to classify spending data by subsector. Spending data may also be classified by other proprietary methods not using MCCs. Generations, if discussed, are defined as follows: 1. Gen Z, born after 1995 2. Younger Millennials: born between 1989-1995 3. Older Millennials: born between 1978-19886 11 March 2024BANK OF AMERICA INSTITUTE4. Gen Xers: born between 1965-1977 5. Baby Boomer: 1946-1964 6. Traditionalists: pre-1946 Any reference to card spending per household on gasoline includes all purchases at gasoline stations and might include purchases of non-gas items. Additional information about the methodology used to aggregate the data is available upon request.\n\n## Contributors\n\nDavid Michael Tinsley Senior Economist, Bank of America Institute Joe Wadford Economist, Bank of America Institute Taylor Bowley Economist, Bank of America Institute Liz Everett Krisberg Head of Bank of America Institute\n\n## Sources\n\nLi Wei Director, Global Risk Analytics Kimberly Warren Director, Global Risk Analytics Ana Maxim Senior Vice President, Consumer and Small Business Mel Roasa Vice President, Digital and MarketingBANK OF AMERICA INSTITUTE11 March 2024 7",
          "relevanceScore": 0.3,
          "documentMetadata": {
            "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/719bfb7c1c0cde3888debd43542aabfe",
            "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1odEo6QRllsURLZRDwHNruCXK9bsWmhtR.pdf",
            "title": "1odEo6QRllsURLZRDwHNruCXK9bsWmhtR",
            "pageIdentifier": "6"
          }
        }
      }
    ],
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Rank generations from highest to lowest based on their spending on holiday items using cards."
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/935c4e1f18ccff2b5fa51d6d00e40dc4",
                  "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1baMNJuizoK7u3P2-gYRwpVz_46-uOhn4.pdf",
                  "title": "1baMNJuizoK7u3P2-gYRwpVz_46-uOhn4",
                  "snippetInfo": [
                    {
                      "snippet": "## Exhibit 6: Credit and debit \u003cb\u003ecard spending\u003c/b\u003e per household on \u003cb\u003eholiday items\u003c/b\u003e by \u003cb\u003egeneration\u003c/b\u003e (index, Aug-Sep average=100 for each year, 7- day moving average) ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/0b4c8cfb6f5ed9ef0df70ffcd79fe2c0",
                  "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1pVkzcMDNAy-p7AlrE0LRlhpbbDzCNndJ.pdf",
                  "title": "1pVkzcMDNAy-p7AlrE0LRlhpbbDzCNndJ",
                  "snippetInfo": [
                    {
                      "snippet": "Consumer \u003cb\u003espending\u003c/b\u003e finished solidly in 2023, \u003cb\u003ewith\u003c/b\u003e total \u003cb\u003ecard spending\u003c/b\u003e per household increasing by 0.2% year-over-year (YoY) in December, according to Bank of ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f8d7887862167c5daf6c7a30e1d464e0",
                  "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1WkoquhDpqHphSnqIVKX45iers7kvmGjZ.pdf",
                  "title": "1WkoquhDpqHphSnqIVKX45iers7kvmGjZ",
                  "snippetInfo": [
                    {
                      "snippet": "This could be due to an increasing customer \u003cb\u003ebase\u003c/b\u003e or inactive customers \u003cb\u003eusing their cards\u003c/b\u003e more frequently. Per household \u003cb\u003ecard spending\u003c/b\u003e growth only looks at ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/719bfb7c1c0cde3888debd43542aabfe",
                  "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1odEo6QRllsURLZRDwHNruCXK9bsWmhtR.pdf",
                  "title": "1odEo6QRllsURLZRDwHNruCXK9bsWmhtR",
                  "snippetInfo": [
                    {
                      "snippet": "This could be due to an increasing customer \u003cb\u003ebase\u003c/b\u003e or inactive customers \u003cb\u003eusing their cards\u003c/b\u003e more frequently. Per household \u003cb\u003ecard spending\u003c/b\u003e growth only looks at ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/c4308d8610bcae3b3d2bb916090ef28b",
                  "uri": "gs://yipeiw_multimodal_0827/rzilleruelo_multimodal_datasets/20240806/Document Understanding Evaluation Dataset/Bank of America/1B3MJqELqk_S26aWb-KnLdSWwmpuB32os.pdf",
                  "title": "1B3MJqELqk_S26aWb-KnLdSWwmpuB32os",
                  "snippetInfo": [
                    {
                      "snippet": "This could be due to an increasing customer \u003cb\u003ebase\u003c/b\u003e or inactive customers \u003cb\u003eusing their cards\u003c/b\u003e more frequently. Per household \u003cb\u003ecard spending\u003c/b\u003e growth only looks at ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ],
    "blobAttachments": [
      {
        "data": {
          "mimeType": "image/png",
          "data": "iVBORw0KGgoAAAANSUhEUgAABcwAAAIjCAIAAACxms+ZAAAgAElEQVR42uzdd1xTV/...Vv/F0NDQ4NOp3+waz94PJ5Op7+/ibro6fRiWPX/AaZ6/dAGW00gAAAAAElFTkSuQmCC"
        },
        "attributionType": "CORPUS"
      }
      ]
  },
  "answerQueryToken": "M8gKCwiPuOe-BhDyhpgIEiQ2N2Q2MWI0Mi0wMDAwLTJjNTgtODFiNy0wODllMDgyY2FkODA"
}

画像は出力の blobAttachments 部分で返されます。回答とともに返される画像は、常に最初の blobAttachments にあります。存在する場合、他の blobAttachments には引用参照で返された画像が含まれます。回答とともに返される画像は、引用とともに返されることもあります（実際には、引用とともに返されることが多いです）。blobAttachmentIndexes は、blobAttachment を回答または引用テキストに関連付けるインデックスです。

返される画像の attributionType は常に CORPUS で、画像がデータストアからのものであることを示します。

フォローアップの質問のコマンド

フォローアップはマルチターンクエリです。フォローアップセッションの最初のクエリの後、後続の「ターン」では以前のやり取りが考慮されます。フォローアップでは、answer メソッドで関連する質問を提示することもできます。ユーザーは、独自のフォローアップの質問を入力する代わりに、提示された質問を選択できます。関連する質問の候補を表示するには、[生成回答] オプションをオンにする必要があります。

前のセクションで説明した回答とフォローアップの機能（引用、フィルタ、セーフサーチ、特定の種類のクエリの無視、プリアンブルを使用した回答のカスタマイズなど）はすべて、フォローアップとともに適用できます。

フォローアップセッションの例

フォローアップ付きセッションの例を次に示します。メキシコでの休暇について調べたいとします。

ターン 1:
- あなた: メキシコで休暇を過ごすのに最適な時期はいつですか？
- フォローアップ付きの回答: メキシコで休暇を過ごすのに最適な時期は乾季です。これは 11 月から 4 月までです。
ターン 2:
- あなた: 為替レートを教えてください。
- フォローアップ付きの回答: 1 米ドルは約 17.65 メキシコペソに相当します。
ターン 3:
- あなた: 12 月の平均気温はどれくらいですか？
- フォローアップ付きの回答: 平均気温は華氏 70～78 度です。カンクンの平均気温は華氏 77 度までです。

フォローアップがないと、「為替レートはいくらですか？」という質問に答えることはできません。通常の検索では、メキシコの為替レートを知りたいことがわからないためです。同様に、フォローアップなしでは、特にメキシコの温度を示すために必要なコンテキストがありません。

セッションについて

エージェント検索でのフォローアップの仕組みを理解するには、セッションについて理解する必要があります。

セッションは、ユーザーが提供したテキストクエリと Agent Search が提供したレスポンスで構成されます。

このようなクエリとレスポンスのペアは、ターンと呼ばれることもあります。上記の例では、2 番目のターンは「為替レートはいくらですか？」と「1 米ドルは約 17.65 メキシコペソに相当します」で構成されています。

セッションはアプリとともに保存されます。アプリでは、セッションはセッションリソースで表されます。

クエリメッセージとレスポンスメッセージに加えて、セッションリソースには次のものも含まれます。

一意の名前（セッション ID）。
状態（進行中または完了）。
ユーザー疑似 ID（ユーザーを追跡する訪問者 ID）。プログラムで割り当てることができます。アプリのユーザーイベントのユーザー疑似 ID にマッピングすると、モデルはユーザーにパーソナライズされた結果を提供できます。
開始時刻と終了時刻。
ターン（クエリと回答のペア）。

始める前に

フォローアップ質問をリクエストするクエリを実行する前に、アプリの [生成された回答] オプションがオンになっていることを確認してください。

セッション情報を保存してレスポンスを取得する

コマンドラインを使用して、検索レスポンスと回答を生成し、各クエリとともにセッションに保存できます。

REST

コマンドラインを使用してセッションを作成し、ユーザーの入力からレスポンスを生成する手順は次のとおりです。

セッションを保存するアプリを指定します。
```
curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions" \
  -d '{
        "userPseudoId": "USER_PSEUDO_ID"
      }'
```
次のように置き換えます。
- PROJECT_ID: 実際の Google Cloud プロジェクト ID。
- APP_ID: Agent Search アプリの ID。
- USER_PSEUDO_ID: ユーザーをトラッキングする一意の仮名化 ID として機能する UTF-8 エンコード文字列。最大長は 128 文字です。このフィールドを使用すると、モデルのパフォーマンスとパーソナライズの品質が向上するため、このフィールドを使用することを強くおすすめします。このフィールドには HTTP Cookie を使用できます。これにより、1 つのデバイス上の訪問者を一意に識別できます。重要な考慮事項は次のとおりです。
  - この ID は、訪問者がウェブサイトにログインまたはログアウトしても変更されません。
  - このフィールドを複数のユーザーに対して同じ ID に設定しないでください。同じユーザー ID を使用すると、異なるユーザーのイベント履歴が統合され、モデルの品質が低下する可能性があります。
  - このフィールドに個人を特定できる情報（PII）を含めることはできません。
  - 特定の検索リクエストまたは閲覧リクエストの場合、このフィールドはユーザーイベントの対応する userPseudoId フィールドにマッピングする必要があります。
  詳細については、userPseudoId をご覧ください。
コマンドの例と結果
```
curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/sessions"
-d '{
"userPseudoId": "test_user"
}'

{
  "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943",
  "state": "IN_PROGRESS",
  "userPseudoId": "test_user",
  "startTime": "2024-09-13T18:47:10.465311Z",
  "endTime": "2024-09-13T18:47:10.465311Z"
}
```
JSON レスポンスの name: フィールドの末尾にある数値をセッション ID としてメモします。この結果の例の場合、ID は 5386462384953257772 です。この ID は次のステップで必要になります。

回答を生成して、アプリのセッションに追加します。

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:answer" \
  -d '{
        "query": { "text": "QUERY"},
        "session": "projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions/SESSION_ID",
          "searchSpec":{ "searchParams": {"filter": "FILTER"} }
}'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
QUERY: 質問または検索クエリを含むフリーテキストの文字列。
SESSION_ID: ステップ 1 で作成したセッションの ID。これは、ステップ 2 でメモした name: フィールドの末尾の数字です。セッションでは、すべてのターンで同じセッション ID を使用します。
FILTER: フィルタ式を使用して検索をフィルタリングするためのテキストフィールド。デフォルト値は空文字列です。フィルタの作成方法は、メタデータを含む非構造化データ、構造化データ、ウェブサイトのデータのいずれがあるかによって異なります。詳細については、構造化データまたは非構造化データのカスタム検索をフィルタするとウェブサイト検索をフィルタするをご覧ください。

コマンドの例と結果

curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:answer"
-d '{
"query": { "text": "Compare bigquery with spanner database?"},
"session":  "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943",
}'
    
{
  "answer": {
    "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943/answers/4861507376861383072",
    "state": "SUCCEEDED",
    "answerText": "BigQuery and Spanner are both powerful tools that can be used together to handle transactional and analytical workloads. Spanner is a fully managed relational database optimized for transactional workloads, while BigQuery is a serverless data warehouse designed for business agility. Spanner provides seamless replication across regions in Google Cloud and processes over 1 billion requests per second at peak. BigQuery analyzes over 110 terabytes of data per second. Users can leverage federated queries to read data from Spanner and write to a native BigQuery table. \n",
    "steps": [
      {
        "state": "SUCCEEDED",
        "description": "Rephrase the query and search.",
        "actions": [
          {
            "searchAction": {
              "query": "Compare bigquery with spanner database?"
            },
            "observation": {
              "searchResults": [
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/ecc0e7547253f4ca3ff3328ce89995af",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/how-spanner-and-bigquery-work-together-handle-transactional-and-analytical-workloads",
                  "title": "How Spanner and BigQuery work together to handle transactional and analytical workloads | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "Using Cloud \u003cb\u003eSpanner\u003c/b\u003e and \u003cb\u003eBigQuery\u003c/b\u003e also allows customers to build their \u003cb\u003edata\u003c/b\u003e clouds using Google Cloud, a unified, open approach to \u003cb\u003edata\u003c/b\u003e-driven transformation ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/d7e238f73608a860e00b752ef80e2941",
                  "uri": "https://cloud.google.com/blog/products/databases/cloud-spanner-gets-stronger-with-bigquery-federated-queries",
                  "title": "Cloud Spanner gets stronger with BigQuery-federated queries | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "As enterprises compete for market share, their need for real-time insights has given rise to increased demand for transactional \u003cb\u003edatabases\u003c/b\u003e to support \u003cb\u003edata\u003c/b\u003e ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/e10a5a3c267dc61579e7c00fefe656eb",
                  "uri": "https://cloud.google.com/blog/topics/developers-practitioners/replicating-cloud-spanner-bigquery-scale",
                  "title": "Replicating from Cloud Spanner to BigQuery at scale | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "... \u003cb\u003eSpanner data\u003c/b\u003e into \u003cb\u003eBigQuery\u003c/b\u003e for analytics. In this post, you will learn how to efficiently use this feature to replicate large tables with high throughput ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                },
                ...
                {
                  "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/8100ad36e1cac149eb9fc180a41d8f25",
                  "uri": "https://cloud.google.com/blog/products/gcp/from-nosql-to-new-sql-how-spanner-became-a-global-mission-critical-database",
                  "title": "How Spanner became a global, mission-critical database | Google Cloud Blog",
                  "snippetInfo": [
                    {
                      "snippet": "... SQL \u003cb\u003evs\u003c/b\u003e. NoSQL dichotomy may no longer be relevant." The \u003cb\u003eSpanner\u003c/b\u003e SQL query processor, while recognizable as a standard implementation, has unique ...",
                      "snippetStatus": "SUCCESS"
                    }
                  ]
                }
              ]
            }
          }
        ]
      }
    ]
  },
  "session": {
    "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943",
    "state": "IN_PROGRESS",
    "userPseudoId": "test_user",
    "turns": [
      {
        "query": {
          "queryId": "projects/123456/locations/global/questions/741830",
          "text": "Compare bigquery with spanner database?"
        },
        "answer": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943/answers/4861507376861383072"
      }
    ],
    "startTime": "2024-09-13T18:47:10.465311Z",
    "endTime": "2024-09-13T18:47:10.465311Z"
  },
  "answerQueryToken": "NMwKDAjFkpK3BhDU24uZAhIkNjZlNDIyZWYtMDAwMC0yMjVmLWIxMmQtZjQwMzA0M2FkYmNj"
}

セッション内の新しいクエリごとにステップ 3 を繰り返します。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine


def create_session(
    project_id: str,
    location: str,
    engine_id: str,
    user_pseudo_id: str,
) -> discoveryengine.Session:
    """Creates a session.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
        user_pseudo_id: A unique identifier for tracking visitors. For example, this
          could be implemented with an HTTP cookie, which should be able to
          uniquely identify a visitor on a single device.
    Returns:
        discoveryengine.Session: The newly created Session.
    """

    client = discoveryengine.SessionServiceClient()

    session = client.create_session(
        # The full resource name of the engine
        parent=f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}",
        session=discoveryengine.Session(user_pseudo_id=user_pseudo_id),
    )

    # Send Session name in `answer_query()`
    print(f"Session: {session.name}")
    return session

データストアからセッションを取得する

次のコマンドは、get メソッドを呼び出してデータストアからセッションを取得する方法を示しています。

REST

データストアからセッションを取得する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X GET -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions/SESSION_ID"

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
SESSION_ID: 取得するセッションの ID。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine


def get_session(
    project_id: str,
    location: str,
    engine_id: str,
    session_id: str,
) -> discoveryengine.Session:
    """Retrieves a session.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
        session_id: The ID of the session.
    """

    client = discoveryengine.SessionServiceClient()

    # The full resource name of the session
    name = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/sessions/{session_id}"

    session = client.get_session(name=name)

    print(f"Session details: {session}")
    return session

アプリからセッションを削除する

次のコマンドは、delete メソッドを呼び出して、データストアからセッションを削除する方法を示しています。

デフォルトでは、60 日を超えるセッションは自動的に削除されます。ただし、特定のセッションを削除する場合（たとえば、機密コンテンツが含まれている場合など）は、この API 呼び出しを使用して削除します。

REST

アプリからセッションを削除するには、次の操作を行います。

次の curl コマンドを実行します。

curl -X DELETE -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions/SESSION_ID"

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
SESSION_ID: 削除するセッションの ID。

コマンドの例と結果

curl -X DELETE -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943"
    
{}

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine


def delete_session(
    project_id: str,
    location: str,
    engine_id: str,
    session_id: str,
) -> None:
    """Deletes a session.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
        session_id: The ID of the session.
    """

    client = discoveryengine.SessionServiceClient()

    # The full resource name of the session
    name = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/sessions/{session_id}"

    client.delete_session(name=name)

    print(f"Session {name} deleted.")

セッションを更新する

セッションを更新する理由はさまざまです。たとえば、次のいずれかを行うためです。

セッションを完了としてマークする
あるセッションのメッセージを別のセッションに統合する
ユーザーの疑似 ID を変更する

次のコマンドは、patch メソッドを呼び出してデータストア内のセッションを更新する方法を示しています。

REST

アプリからセッションを更新する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X PATCH \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions/SESSION_ID?updateMask=state" \
  -d '{
        "state": "NEW_STATE"
      }'

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
SESSION_ID: 更新するセッションの ID。
NEW_STATE: 状態の新しい値（IN_PROGRESS など）。

コマンドの例と結果

curl -X PATCH -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943?updateMask=state"
-d '{
"state": "IN_PROGRESS"
}'
    
{
  "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943",
  "state": "IN_PROGRESS",
  "userPseudoId": "test_user",
  "turns": [
    {
      "query": {
        "queryId": "projects/123456/locations/global/questions/741830",
        "text": "Compare bigquery with spanner database?"
      },
      "answer": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943/answers/4861507376861383072"
    }
  ],
  "startTime": "2024-09-13T18:47:10.465311Z",
  "endTime": "2024-09-13T18:49:41.579151Z"
}

この例では、セッションの状態をオープン（進行中）に変更します。同様のパターンで userPseudoId を更新します。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine
from google.protobuf import field_mask_pb2


def update_session(
    project_id: str,
    location: str,
    engine_id: str,
    session_id: str,
) -> discoveryengine.Session:
    """Updates a session.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
        session_id: The ID of the session.
    Returns:
        discoveryengine.Session: The updated Session.
    """
    client = discoveryengine.SessionServiceClient()

    # The full resource name of the session
    name = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}/sessions/{session_id}"

    session = discoveryengine.Session(
        name=name,
        state=discoveryengine.Session.State.IN_PROGRESS,  # Options: IN_PROGRESS, STATE_UNSPECIFIED
    )

    # Fields to Update
    update_mask = field_mask_pb2.FieldMask(paths=["state"])

    session = client.update_session(session=session, update_mask=update_mask)
    print(f"Updated session: {session.name}")
    return session

すべてのセッションを一覧表示する

次のコマンドは、list メソッドを呼び出して、データストア内のセッションを一覧表示する方法を示しています。

REST

アプリのセッションを一覧表示する手順は次のとおりです。

次の curl コマンドを実行します。

curl -X GET \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions"

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。

コマンドの例と結果

curl -X GET -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/sessions"
    
{
  "sessions": [
    {
      "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10000135306311111817",
      "state": "IN_PROGRESS",
      "turns": [
        {
          "query": {
            "queryId": "projects/123456/locations/global/questions/10000135306311114276",
            "text": "bugs reported by tiktok on grounding"
          }
        }
      ],
      "startTime": "2024-09-03T00:38:40.338623Z",
      "endTime": "2024-09-03T00:38:40.338623Z"
    },
    {
      "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10000827040519035859",
      "state": "IN_PROGRESS",
      "turns": [
        {
          "query": {
            "queryId": "projects/123456/locations/global/questions/10000827040519033518",
            "text": "GDM models"
          }
        }
      ],
      "startTime": "2024-07-19T15:53:06.521775Z"
    },
    {
      "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10003910515245149877",
      "state": "IN_PROGRESS",
      "turns": [
        {
          "query": {
            "queryId": "projects/123456/locations/global/questions/10003910515245148378",
            "text": "gyorgyattila"
          },
          "answer": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10003910515245149877/answers/17036357111873257990"
        }
      ],
      "startTime": "2024-08-08T11:40:04.632463Z",
      "endTime": "2024-08-08T11:40:04.632463Z"
    },
...
    {
      "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10198752942940073431",
      "state": "IN_PROGRESS",
      "turns": [
        {
          "query": {
            "queryId": "projects/123456/locations/global/questions/10198752942940071818",
            "text": "hello"
          },
          "answer": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/10198752942940073431/answers/13411441797796265380"
        }
      ],
      "startTime": "2024-08-14T17:30:21.203439Z",
      "endTime": "2024-08-14T17:30:21.203439Z"
    }
  ],
  "nextPageToken": "IDEDgIwL_vuieLC"
}

レスポンスには、セッションのリストと nextPageToken が含まれます。nextPageToken が返されない場合、一覧表示するセッションはありません。デフォルトのページサイズは 50 です。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine


def list_sessions(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.ListSessionsResponse:
    """Lists all sessions associated with a data store.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
    Returns:
        discoveryengine.ListSessionsResponse: The list of sessions.
    """

    client = discoveryengine.SessionServiceClient()

    # The full resource name of the engine
    parent = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}"

    response = client.list_sessions(
        request=discoveryengine.ListSessionsRequest(
            parent=parent,
            filter='state="IN_PROGRESS"',  # Optional: Filter requests by userPseudoId or state
            order_by="update_time",  # Optional: Sort results
        )
    )

    print("Sessions:")
    for session in response.sessions:
        print(session)

    return response

ユーザーのセッションを一覧表示する

次のコマンドは、list メソッドを呼び出して、ユーザーまたはサイト訪問者に関連付けられたセッションを一覧表示する方法を示しています。

REST

ユーザーまたは訪問者に関連付けられているセッションを一覧表示するには、次の操作を行います。

次の curl コマンドを実行します。

curl -X GET \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions?filter=userPseudoId=USER_PSEUDO_ID"

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
USER_PSEUDO_ID: セッションを一覧表示するユーザーの疑似 ID。

コマンドの例と結果

curl -X GET -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/sessions?filter=userPseudoId=test_user"
    
{
  "sessions": [
    {
      "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943",
      "state": "IN_PROGRESS",
      "userPseudoId": "test_user",
      "turns": [
        {
          "query": {
            "queryId": "projects/123456/locations/global/questions/741830",
            "text": "Compare bigquery with spanner database?"
          },
          "answer": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943/answers/4861507376861383072"
        }
      ],
      "startTime": "2024-09-13T18:47:10.465311Z",
      "endTime": "2024-09-13T18:49:41.579151Z"
    }
  ]
}

この例では、test_user に関連付けられているセッションが 1 つあります。セッション内のクエリと回答が一覧表示されます。

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine


def list_sessions(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.ListSessionsResponse:
    """Lists all sessions associated with a data store.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
    Returns:
        discoveryengine.ListSessionsResponse: The list of sessions.
    """

    client = discoveryengine.SessionServiceClient()

    # The full resource name of the engine
    parent = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}"

    response = client.list_sessions(
        request=discoveryengine.ListSessionsRequest(
            parent=parent,
            filter='state="IN_PROGRESS"',  # Optional: Filter requests by userPseudoId or state
            order_by="update_time",  # Optional: Sort results
        )
    )

    print("Sessions:")
    for session in response.sessions:
        print(session)

    return response

セッションのユーザーと状態を一覧表示する

次のコマンドは、list メソッドを呼び出して、特定のユーザーの特定の状態のセッションを一覧表示する方法を示しています。

REST

オープンまたはクローズしていて、特定のユーザーまたは訪問者に関連付けられているユーザーのセッションを一覧表示するには、次の操作を行います。

次の curl コマンドを実行します。

curl -X GET -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/sessions?filter=userPseudoId=USER_PSEUDO_ID%20AND%20state=STATE"

次のように置き換えます。

PROJECT_ID: 実際の Google Cloud プロジェクト ID。
APP_ID: Agent Search アプリの ID。
USER_PSEUDO_ID: セッションを一覧表示するユーザーの疑似 ID。
STATE: セッションの状態: STATE_UNSPECIFIED（クローズ、または不明）または IN_PROGRESS（オープン）。

コマンドの例と結果

 
curl -X GET -H "Authorization: Bearer $(gcloud auth print-access-token)"
-H "Content-Type: application/json"
"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/engines/my-app/sessions?filter=userPseudoId=test_user%20AND%20state=IN_PROGRESS"
    
{
  "sessions": [
    {
      "name": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943",
      "state": "IN_PROGRESS",
      "userPseudoId": "test_user",
      "turns": [
        {
          "query": {
            "queryId": "projects/123456/locations/global/questions/741830",
            "text": "Compare bigquery with spanner database?"
          },
          "answer": "projects/123456/locations/global/collections/default_collection/engines/my-app/sessions/16002628354770206943/answers/4861507376861383072"
        }
      ],
      "startTime": "2024-09-13T18:47:10.465311Z",
      "endTime": "2024-09-13T18:49:41.579151Z"
    }
  ]
}

Python

詳細については、Agent Search Python API リファレンスドキュメントをご覧ください。

from google.cloud import discoveryengine_v1 as discoveryengine


def list_sessions(
    project_id: str,
    location: str,
    engine_id: str,
) -> discoveryengine.ListSessionsResponse:
    """Lists all sessions associated with a data store.

    Args:
        project_id: The ID of your Google Cloud project.
        location: The location of the app.
        engine_id: The ID of the app.
    Returns:
        discoveryengine.ListSessionsResponse: The list of sessions.
    """

    client = discoveryengine.SessionServiceClient()

    # The full resource name of the engine
    parent = f"projects/{project_id}/locations/{location}/collections/default_collection/engines/{engine_id}"

    response = client.list_sessions(
        request=discoveryengine.ListSessionsRequest(
            parent=parent,
            filter='state="IN_PROGRESS"',  # Optional: Filter requests by userPseudoId or state
            order_by="update_time",  # Optional: Sort results
        )
    )

    print("Sessions:")
    for session in response.sessions:
        print(session)

    return response

回答とフォローアップを取得する コレクションでコンテンツを整理 必要に応じて、コンテンツの保存と分類を行います。

answer メソッドの機能

回答と検索を使用するタイミング

クエリフェーズの機能

クエリの言い換え

クエリ分類

検索フェーズの機能

回答フェーズの機能

始める前に

検索と回答（基本）

REST

コマンドの例と部分的な結果

Python

クエリフェーズのコマンド

検索と回答（言い換えが無効）

REST

コマンドの例と部分的な結果

Python

検索と回答（最大ステップ数を指定）

REST

コマンドの例

Python

クエリ分類による検索と回答

REST

コマンドの例と部分的な結果

Python

検索フェーズのコマンド: 検索して検索結果のオプション付きで回答する

REST

コマンドの例と結果

Python

回答フェーズのコマンド

敵対的クエリと回答を求めていないクエリを無視する

REST

コマンドの例と部分的な結果

Python

関連する回答のみを表示する

REST

コマンドの例と結果

Python

グラウンディング サポート スコアを返す

REST

コマンドの例と部分的な結果

根拠のある回答のみを表示する

REST

コマンドの例と部分的な結果

回答モデルを指定する

REST

コマンドの例と部分的な結果

Python

カスタム プリアンブルを指定する

REST

コマンドの例と結果

Python

関連する質問を取得する

始める前に

手順

REST

コマンドの例と部分的な結果

引用を含める

REST

コマンドの例と部分的な結果

Python

回答の言語コードを設定する

REST

コマンドの例と結果

Python

回答をパーソナライズする

コマンドの例と結果

回答のグラフを生成する

制約事項

一般的な障害シナリオ

始める前に

手順

REST

コマンドの例と部分的な結果

データストアから既存の画像を取得する

制限事項

手順

REST

コマンドの例と部分的な結果

回答とフォローアップを取得する

グラウンディングサポートスコアを返す

カスタムプリアンブルを指定する

フォローアップセッションの例