Create a Vertex Data Store

This is a prerequisite step before you create a データ ストアに基づく モデル接続 for using the Google Vertex AI RAG capability.

You would first create a bucket in Google Cloud, and then create a Vertex Data Store in the エージェント ビルダー. We recommend these steps as an example.

Create a bucket in Google Cloud

Before creating a Data Store you would create a bucket in Google Cloud and upload your data to it.

Create a bucket in Google Cloud

  1. Log in to your Google Cloud account and navigate to Cloud Storage and click Create Bucket.
  2. Provide a name for your bucket, keep the remaining parameters as per default settings and click Continue.
  3. Next, upload your documents into this new bucket. This could comprise of any document format the Vertex Data Store supports, such as: PDF, HTML, TXT, and others.

Create a Vertex Data Store

Once you have completed creating a bucket, you would navigate to the Agent Builder > Data Stores tab to create a data store.

  1. In the Create Data Store > Select a data source screen, select Cloud Storage.
    Note: You have already created a bucket in Cloud Storage in the earlier steps.

    Create a Data Store in エージェント ビルダー

  2. Select Import data from Cloud Storage > What kind of data are you importing? > Unstructured documents (PDF, HTML, TXT, and more).
  3. Keep all other parameters as per default selection, but specify the folder.
  4. Click Folder and select the bucket you created earlier, and click Continue.

    Configure a Data Store

  5. In the Configure your data store screen, keep the default options for Location of your data store field.
  6. Provide a name for the Your data store name field.
    Note: The data store comprises of vectorized data that is chunked and encoded, and used for RAG solutions. If you want to chunk your content then you would configure it at this point.
  7. Expand the Document Processing Options section.

    Configure document parsing for the Data Store

  8. In the Document parsing section, we recommend selecting Layout Parser as it works well for most document types.
    Note: See Parse and chunk documents .
  9. Selecting the Layout Parser option enables Document chunking.
  10. Next, check the Include ancestor headings in chunks option as this helps return a more comprehensive response picked from multiple chunks with overlapping data. Checking this option is optional, as per your requirement.
  11. Click Create to complete configuring the data store.

Google データ ソース summary

In the エージェント ビルダー, navigate to Apps > Data Stores and click the data store you created. You would see a summary screen of your data store with information that you would reference when creating a データ ストアに基づく モデル接続 in AI エージェント Studio.

Data Store summary view

Project ID
This is the ID of the project as it appears in the Google データ ソース. You would require this value when creating a データ ストアに基づく モデル接続 in AI エージェント Studio. You would find this ID when you select your data store.

Where to locate your Project ID

Data store ID
You would use this ID for defining the authentication details when creating a データ ストアに基づく モデル接続 in AI エージェント Studio.
Region
Displays the region the data store is deployed in.

Select a region from the drop-down list to connect for authenticating the モデル接続. You can also add your own region that you configured when you created a data source in エージェント ビルダー. See Create a data store .

Connected apps
Shows the Apps the data store is connected to.
Note: You can create an App in the Google データ ソース and connect it to the data store you created, if you want to test the data source directly in the Google データ ソース.

Authentication details needed for creating Google Vertex AI モデル接続 in AI エージェント Studio

After creating a data store, uploading documents to it takes time. Click on Data Store > Activity to view the status of the upload.