Grounding with Vertex AI Search

This page explains how you can ground responses by using your data from Vertex AI Search.

Grounding Gemini to your data

If you want to do retrieval-augmented generation (RAG), connect your model to your website data or your sets of documents, then use Grounding with Vertex AI Search .

Grounding to your data supports a maximum of 10 Vertex AI Search data sources and can be combined with Grounding with Google Search .

Supported models

This section lists the models that support grounding with your data.

Gemini 2.5 Flash (Preview)
Gemini 2.5 Flash-Lite (Preview)
Gemini 2.5 Flash-Lite
Gemini 2.5 Flash with Live API native audio (Preview)
Gemini 2.0 Flash with Live API (Preview)
Gemini 2.5 Pro
Gemini 2.5 Flash
Gemini 2.0 Flash

Prerequisites

Before you can ground model output to your data, do the following:

In the Google Cloud console, go to the IAMpage, and search for the discoveryengine.servingConfigs.search permission, which is required for the grounding service to work.

Go to IAM
Enable AI Applications and activate the API.
Create a AI Applications data source and application.

See the Introduction to Vertex AI Search for more.

Enable AI Applications

To use Vertex AI Search to ground your responses, you must activate the Vertex AI Search service by following these steps:

In the Google Cloud console, go to the AI Applicationspage.

Go to AI Applications
Optional: Review the terms for data use .

AI Applications is available in the global location or the eu and us multi-region. To learn more, see AI Applications locations .

Create a data store in AI Applications

To create a data store in AI Applications, you can choose to ground with website data or documents.

Website

Open the Create Data Store page from the Google Cloud console.
In Website Contentbox, click Select .
Specify the websites for your data storepane displays.
If Advanced website indexingisn't checked, then select the Advanced website indexingcheckbox to turn it on.
Configure your data storepane displays.
In the Specify URL patterns to indexsection, do the following:
- Add URLs for Sites to include.
- Optional: Add URLs for Sites to exclude.
Click Continue.
In the Configure your data storepane,
1. Select a value from the Location of your data storelist.
2. Enter a name in the Your data store namefield. The ID is generated. Use this ID when you generate your grounded responses with your data store. For more information, see Generate grounded responses with your data store .
3. Click Create.

Documents

Open the Create Data Store page from the Google Cloud console.
In Cloud Storagebox, click Select .
Import data from Cloud Storagepane displays.
In the Unstructured documents (PDF, HTML, TXT and more)section, select Unstructured documents (PDF, HTML, TXT and more).
Select a Synchronization frequencyoption.
Select a Select a folder or a file you want to importoption, and enter the path in the field.
Click Continue.
Configure your data storepane displays.
In the Configure your data storepane,
1. Select a value from the Location of your data storelist.
2. Enter a name in the Your data store namefield. The ID is generated.
3. To select parsing and chunking options for your documents, expand the Document Processing Optionssection. For more information about different parsers, see Parse documents .
4. Click Create.
Click Create.

Generate grounded responses with your data store

Use the following instructions to ground a model with your data. A maximum of 10 data stores is supported.

If you don't know your data store ID, follow these steps:

In the Google Cloud console, go to the AI Applicationspage and in the navigation menu, click Data stores.

Go to the Data stores page
Click the name of your data store.
On the Data page for your data store, get the data store ID.

Console

To ground your model output to AI Applications by using Vertex AI Studio in the Google Cloud console, follow these steps:

In the Google Cloud console, go to the Vertex AI Studio Freeform page.
Go to Freeform
To turn on grounding, click the Grounding: your datatoggle.
Click Customize.
1. Select Vertex AI Searchas your source.
2. Using this path format, replace your data store's Project ID and the ID of the data store:
  
  projects/ PROJECT_ID /locations/global/collections/default_collection/dataStores/ DATASTORE_ID .
Click Save.
Enter your prompt in the text box, and click Submit.

Your prompt responses are grounded to AI Applications.

Python

Install

pip install --upgrade google-genai

To learn more, see the SDK reference documentation .

Set environment variables to use the Gen AI SDK with Vertex AI:

 # Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values 
 # with appropriate values for your project. 
 export 
  
 GOOGLE_CLOUD_PROJECT 
 = 
 GOOGLE_CLOUD_PROJECT 
 export 
  
 GOOGLE_CLOUD_LOCATION 
 = 
 global 
 export 
  
 GOOGLE_GENAI_USE_VERTEXAI 
 = 
True

  from 
  
 google 
  
 import 
 genai 
 from 
  
 google.genai.types 
  
 import 
 ( 
 GenerateContentConfig 
 , 
 VertexAISearch 
 , 
 Retrieval 
 , 
 Tool 
 , 
 HttpOptions 
 , 
 ) 
 client 
 = 
 genai 
 . 
 Client 
 ( 
 http_options 
 = 
 HttpOptions 
 ( 
 api_version 
 = 
 "v1" 
 )) 
 # Replace with your Vertex AI Search data store details 
 DATASTORE_PATH 
 = 
 "projects/ PROJECT_ID 
/locations/global/collections/default_collection/dataStores/DATASTORE_ID" 
 tool 
 = 
 Tool 
 ( 
 retrieval 
 = 
 Retrieval 
 ( 
 vertex_ai_search 
 = 
 VertexAISearch 
 ( 
 datastore 
 = 
 DATASTORE_PATH 
 ) 
 ) 
 ) 
 response 
 = 
 client 
 . 
 models 
 . 
 generate_content 
 ( 
 model 
 = 
 "gemini-2.5-flash" 
 , 
 # Or another supported model 
 contents 
 = 
 "What information can you find about topic X in the provided documents?" 
 , 
 # Your query 
 config 
 = 
 GenerateContentConfig 
 ( 
 tools 
 = 
 [ 
 tool 
 ], 
 ), 
 ) 
 print 
 ( 
 response 
 . 
 text 
 )

REST

To test a text prompt by using the Vertex AI API, send a POST request to the publisher model endpoint.

Before using any of the request data, make the following replacements:

LOCATION : The region to process the request. To use the global endpoint , exclude the location from the endpoint name, and configure the location of the resource to global .
PROJECT_ID : Your project ID .
MODEL_ID : The model ID of the multimodal model.
PROMPT : The prompt to send to the model.

HTTP method and URL:

POST https:// LOCATION 
-aiplatform.googleapis.com/v1beta1/projects/ PROJECT_ID 
/locations/ LOCATION 
/publishers/google/models/ MODEL_ID 
:generateContent

Request JSON body:

{
  "contents": [{
    "role": "user",
    "parts": [{
      "text": " PROMPT 
"
    }]
  }],
  "tools": [{
    "retrieval": {
      "vertexAiSearch": {
        "datastore": projects/ PROJECT_ID 
/locations/global/collections/default_collection/dataStores/ DATASTORE_ID 
}
    }
  }],
  "model": "projects/ PROJECT_ID 
/locations/ LOCATION 
/publishers/google/models/ MODEL_ID 
"
}

To send your request, expand one of these options:

curl (Linux, macOS, or Cloud Shell)

Note: The following command assumes that you have logged in to the gcloud CLI with your user account by running gcloud init or gcloud auth login , or by using Cloud Shell , which automatically logs you into the gcloud CLI . You can check the currently active account by running gcloud auth list .

Save the request body in a file named request.json , and execute the following command:

curl -X POST \
-H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json; charset=utf-8" \
-d @request.json \
"https:// LOCATION 
-aiplatform.googleapis.com/v1beta1/projects/ PROJECT_ID 
/locations/ LOCATION 
/publishers/google/models/ MODEL_ID 
:generateContent"

PowerShell (Windows)

Note: The following command assumes that you have logged in to the gcloud CLI with your user account by running gcloud init or gcloud auth login . You can check the currently active account by running gcloud auth list .

Save the request body in a file named request.json , and execute the following command:

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
-Method POST `
-Headers $headers `
-ContentType: "application/json; charset=utf-8" `
-InFile request.json `
-Uri "https:// LOCATION 
-aiplatform.googleapis.com/v1beta1/projects/ PROJECT_ID 
/locations/ LOCATION 
/publishers/google/models/ MODEL_ID 
:generateContent" | Select-Object -Expand Content

You should receive a JSON response similar to the following:

{
  "candidates": [
    {
      "content": {
        "role": "model",
        "parts": [
          {
            "text": "You can make an appointment on the website https://dmv.gov/"
          }
        ]
      },
      "finishReason": "STOP",
      "safetyRatings": [
        "..."
      ],
      "groundingMetadata": {
        "retrievalQueries": [
          "How to make appointment to renew driving license?"
        ],
        "groundingChunks": [
          {
            "retrievedContext": {
              "uri": "https://vertexaisearch.cloud.google.com/grounding-api-redirect/AXiHM.....QTN92V5ePQ==",
              "title": "dmv"
            }
          }
        ],
        "groundingSupport": [
          {
            "segment": {
              "startIndex": 25,
              "endIndex": 147
            },
            "segment_text": "ipsum lorem ...",
            "supportChunkIndices": [1, 2],
            "confidenceScore": [0.9541752, 0.97726375]
          },
          {
            "segment": {
              "startIndex": 294,
              "endIndex": 439
            },
            "segment_text": "ipsum lorem ...",
            "supportChunkIndices": [1],
            "confidenceScore": [0.9541752, 0.9325467]
          }
        ]
      }
    }
  ],
  "usageMetadata": {
    "..."
  }
}

Understand your response

The response from both APIs include the LLM-generated text, which is called a candidate . If your model prompt successfully grounds to your data source, then the responses include grounding metadata, which identifies the parts of the response that were derived from your data. However, there are several reasons this metadata may not be provided, and the prompt response won't be grounded. These reasons include low-source relevance or incomplete information within the model's response.

The following is a breakdown of the output data:

Role: Indicates the sender of the grounded answer. Because the response always contains grounded text, the role is always model .
Text: The grounded answer generated by the LLM.
Grounding metadata: Information about the grounding source, which contains the following elements:
- Grounding chunks: A list of results from your index that support the answer.
- Grounding supports: Information about a specific claim within the answer that can be used to show citations:
- Segment: The part of the model's answer that is substantiated by a grounding chunk.
- Grounding chunk index: The index of the grounding chunks in the grounding chunks list that corresponds to this claim.
- Confidence scores: A number from 0 to 1 that indicates how grounded the claim is in the provided set of grounding chunks. Not available for Gemini 2.5 and later.

What's next

To learn how to send chat prompt requests, see Multiturn chat .
To learn about responsible AI best practices and Vertex AI's safety filters, see Safety best practices .