Grounding API

In generative AI, grounding is the ability to connect model output to verifiable sources of information. If you provide models with access to specific data sources, then grounding tethers their output to these data and reduces the chances of inventing content.

With Vertex AI, you can ground model outputs in the following ways:

  • Ground with Google Search - ground a model with publicly available web data.
  • Ground to your own data - ground a model with your own data from Vertex AI Search as a data store.

For more information about grounding, see Grounding overview.

Supported Models:

Model Version
Gemini 1.0 Pro with text input only gemini-1.0-pro-001
gemini-1.0-pro-002

Limitations

  • Grounding only supports data sources that are in English, Spanish, and Japanese.
  • Grounding is only available for text requests.

Example syntax

Syntax to ground a model.

curl

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \

http://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/publishers/google/models/${MODEL_ID}:generateContent \
  -d '{
    "contents": [{
      ...
    }],
    "tools": [{
      "retrieval": {
      "googleSearchRetrieval": {}
        }
    }],
    "model": ""
  }'

Parameter list

See examples for implementation details.

GoogleSearchRetrieval

Ground the response with public data.

Parameters

google_search_retrieval

Required: Object

Ground with publicly available web data.

Retrieval

Ground the response with private data from Vertex AI Search as a data store. Defines a retrieval tool that the model can call to access external knowledge.

Parameters

source

Required: VertexAISearch

Ground with Vertex AI Search data sources.

VertexAISearch

Parameters

datastore

Required: string

Fully-qualified data store resource ID from Vertex AI Search, in the following format: projects/{project}/locations/{location}/collections/default_collection/dataStores/{datastore}

Examples

Ground response on public web data using Google Search

Ground the response with Google Search public data. Include the google_search_retrieval tool in the request. No additional parameters are required.

REST

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

  • LOCATION: The region to process the request.
  • PROJECT_ID: Your project ID.
  • MODEL_ID: The model ID of the multimodal model.
  • TEXT: The text instructions to include in the prompt.

HTTP method and URL: 

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

Request JSON body: 

{
  "contents": [{
    "role": "user",
    "parts": [{
      "text": "TEXT"
    }]
  }],
  "tools": [{
    "googleSearchRetrieval": {}
  }],
  "model": "projects/PROJECT_ID/locations/LOCATION/publishers/google/models/MODEL_ID"
}

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

{
   "candidates": [
     {
       "content": {
         "role": "model",
         "parts": [
           {
             "text": "Chicago weather changes rapidly, so layers let you adjust easily. Consider a base layer, a warm mid-layer (sweater-fleece), and a weatherproof outer layer."
           }
         ]
       },
       "finishReason": "STOP",
       "safetyRatings":[
       "..."
    ],
       "groundingMetadata": {
         "webSearchQueries": [
           "What's the weather in Chicago this weekend?"
         ],
         "searchEntryPoint": {
            "renderedContent": "....................."
         }
       }
     }
   ],
   "usageMetadata": { "..."
   }
 }

Python

To learn how to install or update the Vertex AI SDK for Python, see Install the Vertex AI SDK for Python. For more information, see the Python API reference documentation. 

import vertexai

from vertexai.preview.generative_models import grounding
from vertexai.generative_models import GenerationConfig, GenerativeModel, Tool

# TODO(developer): Update and un-comment below line
# project_id = "PROJECT_ID"

vertexai.init(project=project_id, location="us-central1")

model = GenerativeModel(model_name="gemini-1.0-pro-002")

# Use Google Search for grounding
tool = Tool.from_google_search_retrieval(grounding.GoogleSearchRetrieval())

prompt = "When is the next total solar eclipse in US?"
response = model.generate_content(
    prompt,
    tools=[tool],
    generation_config=GenerationConfig(
        temperature=0.0,
    ),
)

print(response)

Node.js

Before trying this sample, follow the Node.js setup instructions in the Vertex AI quickstart using client libraries. For more information, see the Vertex AI Node.js API reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment. 

const {
  VertexAI,
  HarmCategory,
  HarmBlockThreshold,
} = require('@google-cloud/vertexai');

/**
 * TODO(developer): Update these variables before running the sample.
 */
async function generateContentWithGoogleSearchGrounding(
  projectId = 'PROJECT_ID',
  location = 'us-central1',
  model = 'gemini-1.0-pro-002'
) {
  // Initialize Vertex with your Cloud project and location
  const vertexAI = new VertexAI({project: projectId, location: location});

  const generativeModelPreview = vertexAI.preview.getGenerativeModel({
    model: model,
    // The following parameters are optional
    // They can also be passed to individual content generation requests
    safetySettings: [
      {
        category: HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
        threshold: HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
      },
    ],
    generationConfig: {maxOutputTokens: 256},
  });

  const googleSearchRetrievalTool = {
    googleSearchRetrieval: {},
  };

  const request = {
    contents: [{role: 'user', parts: [{text: 'Why is the sky blue?'}]}],
    tools: [googleSearchRetrievalTool],
  };

  const result = await generativeModelPreview.generateContent(request);
  const response = result.response;
  const groundingMetadata = response.candidates[0].groundingMetadata;
  console.log(
    'Response: ',
    JSON.stringify(response.candidates[0].content.parts[0].text)
  );
  console.log('GroundingMetadata is: ', JSON.stringify(groundingMetadata));
}

Ground response on private data using Vertex AI Search

Ground the response with data from a Vertex AI Search data store. For more information, see Vertex AI Agent Builder.

REST

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

  • LOCATION: The region to process the request.
  • PROJECT_ID: Your project ID.
  • MODEL_ID: The model ID of the multimodal model.
  • TEXT: The text instructions to include in the prompt.

HTTP method and URL: 

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

Request JSON body: 

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

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

{
   "candidates": [
     {
       "content": {
         "role": "model",
         "parts": [
           {
             "text": "You can make an appointment on the website http://dmv.gov/"
           }
         ]
       },
       "finishReason": "STOP",
       "safetyRatings":[
       "..."
    ],
       "groundingMetadata": {
         "retrievalQueries": [
           "How to make appointment to renew driving license?"
         ]
       }
     }
   ],
   "usageMetadata": { "..."
   }
 }

Python

To learn how to install or update the Vertex AI SDK for Python, see Install the Vertex AI SDK for Python. For more information, see the Python API reference documentation. 

import vertexai

from vertexai.preview.generative_models import grounding
from vertexai.generative_models import GenerationConfig, GenerativeModel, Tool

# TODO(developer): Update and un-comment below line
# project_id = "PROJECT_ID"

vertexai.init(project=project_id, location="us-central1")

model = GenerativeModel(model_name="gemini-1.0-pro-002")

# Use Vertex AI Search data store
# Format: projects/{project_id}/locations/{location}/collections/default_collection/dataStores/{data_store_id}
tool = Tool.from_retrieval(
    grounding.Retrieval(grounding.VertexAISearch(datastore=data_store_path))
)

prompt = "How do I make an appointment to renew my driver's license?"
response = model.generate_content(
    prompt,
    tools=[tool],
    generation_config=GenerationConfig(
        temperature=0.0,
    ),
)

print(response)

Node.js

Before trying this sample, follow the Node.js setup instructions in the Vertex AI quickstart using client libraries. For more information, see the Vertex AI Node.js API reference documentation.

To authenticate to Vertex AI, set up Application Default Credentials. For more information, see Set up authentication for a local development environment. 

const {
  VertexAI,
  HarmCategory,
  HarmBlockThreshold,
} = require('@google-cloud/vertexai');

/**
 * TODO(developer): Update these variables before running the sample.
 */
async function generateContentWithVertexAISearchGrounding(
  projectId = 'PROJECT_ID',
  location = 'us-central1',
  model = 'gemini-1.0-pro-002',
  dataStoreId = 'DATASTORE_ID'
) {
  // Initialize Vertex with your Cloud project and location
  const vertexAI = new VertexAI({project: projectId, location: location});

  const generativeModelPreview = vertexAI.preview.getGenerativeModel({
    model: model,
    // The following parameters are optional
    // They can also be passed to individual content generation requests
    safetySettings: [
      {
        category: HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
        threshold: HarmBlockThreshold.BLOCK_MEDIUM_AND_ABOVE,
      },
    ],
    generationConfig: {maxOutputTokens: 256},
  });

  const vertexAIRetrievalTool = {
    retrieval: {
      vertexAiSearch: {
        datastore: `projects/${projectId}/locations/global/collections/default_collection/dataStores/${dataStoreId}`,
      },
      disableAttribution: false,
    },
  };

  const request = {
    contents: [{role: 'user', parts: [{text: 'Why is the sky blue?'}]}],
    tools: [vertexAIRetrievalTool],
  };

  const result = await generativeModelPreview.generateContent(request);
  const response = result.response;
  const groundingMetadata = response.candidates[0];
  console.log('Response: ', JSON.stringify(response.candidates[0]));
  console.log('GroundingMetadata is: ', JSON.stringify(groundingMetadata));
}

What's next

For detailed documentation, see the following: