Skip to main content

Amazon Kendra Integration

This guide explains how to configure the display of reference sources when using the RAG assistant feature with Kendra in AI-Starter. This applies to both Azure GPT-4 RAG and AWS Claude 2.1 RAG.

Reference Display Options​

AI-Starter offers three display methods to choose from:

  1. Display all references as links
  2. Link to URLs specified in S3 document metadata
  3. Do not display links (default)

Below, we explain the details and setup methods for each display option.

1. Display all references as links​

Overview

This setting allows users to directly access files in the S3 bucket, which is Kendra's data source, through the application.

Notes

  • Currently, accessible file formats are limited to PDF, HTML, JSON, XML, CSV, Markdown, and text files.

Setup Method

  1. Open the AWS Secrets Manager service page in the AWS Management Console.

  2. Open the cmais-prd/customize secret.

  3. In the "Overview" tab, under "Secret value," click "Retrieve secret value."

  4. Click "Edit."

  5. Enter kendra_doc_link as the key and ALL as the value, then click "Save."

  6. Open the S3 service page.

  7. Display the S3 bucket referenced as Kendra's data source, select the "Properties" tab, and note the "Amazon Resource Name (ARN)."

  8. Open the IAM service page and go to the "Roles" page.

  9. Search for and display cmais-prd-ar-instance-role.

  10. In the "Permissions" tab, click "Add permissions" and choose "Create inline policy."

  11. In the "Policy editor" tab, select "JSON." Paste the following JSON. Enter the ARN of the S3 bucket you noted earlier for Resource. After completing, click "Next."

    {
      "Version": "2012-10-17",
      "Statement": [
        {
          "Sid": "Statement1",
          "Action": "s3:GetObject",
          "Resource": ["{S3 Bucket ARN1}/*", "{S3 Bucket ARN2}/*"],
          "Effect": "Allow"
        }
      ]
    }

  12. Enter ShowDataSourcePermission as the policy name and click "Create policy."

  13. (Only if you updated the secret, perform the following steps)

  14. Open the App Runner service page.

  15. Select cmais-prd-ar-ai-starter.

  16. Click "Deploy."

2. Link to URLs specified in S3 document metadata​

Overview

This setting displays specific attribute values in the metadata set for each document in the S3 bucket, which is Kendra's data source, as URLs.

Setup Method

  1. Create metadata for each S3 document.

    • For details, refer to Amazon S3 document metadata - Amazon Kendra.

    • Create metadata in a JSON file.

      • The file name should be the target document name with .metadata.json appended. For example, the metadata file name for a document named test.txt would be test.txt.metadata.json.

      • The file content should be written as follows. Replace http://example.com with the URL you want to reference.

        {
          "Attributes": {
            "custom_url": "http://example.com"
          }
        }

  2. Open the S3 service page in the AWS Management Console.

  3. Place the created metadata files in the S3 bucket referenced as Kendra's data source.

  4. Open the Kendra service page.

  5. Open the Index that AI-Starter is linked to.

  6. Open "Facet definition."

  7. Click the "Add field" button and add a field with the following settings:

    • Field name: custom_url
    • Date type: String
    • Displayable: Check the box

  8. Open "Data sources."

  9. Check the data source linked to AI-Starter and execute "Sync."

  10. Open the AWS Secrets Manager service page.

  11. Open the cmais-prd/customize secret.

  12. In the "Overview" tab, under "Secret value," click "Retrieve secret value."

  13. Click "Edit."

  14. Enter kendra_doc_link as the key and custom_url as the value, then click "Save."

  15. (Only if you updated the secret, perform the following steps)

  16. Open the App Runner service page.

  17. Select cmais-prd-ar-ai-starter.

  18. Click "Deploy."

3. Do not display links (default)​

Overview

With this setting, users can only see the titles of reference sources and cannot access files or URLs.

Setup Method

  1. Open the AWS Secrets Manager service page in the AWS Management Console.
  2. Open the cmais-prd/customize secret.
  3. In the "Overview" tab, under "Secret value," click "Retrieve secret value."
  4. Click "Edit."
  5. Enter kendra_doc_link as the key and OFF as the value, then click "Save."
  6. (Only if you updated the secret, perform the following steps)
  7. Open the App Runner service page.
  8. Select cmais-prd-ar-ai-starter.
  9. Click "Deploy."