Create Grounded Model connections with Amazon Bedrock RAG capability

Create Grounded by knowledge base Model connections using the native RAG (Retrieval Augmented Generation) capability from Amazon Bedrock to generate accurate and contextually relevant information referenced from Amazon Knowledge Base.

We are now offering you the option to create Model connections using the Amazon Bedrock-RAG capability. A search query on RAG retrieves relevant chunks of content from large-data-sets which are relevant and accurate to the provided context. After retrieving the relevant information, the model uses it to generate a response.

A few things to consider

  • To use the Amazon Bedrock-RAG capability, you would first create a Knowledge Base in Amazon Bedrock before creating the Grounded by knowledge base Model connection.

    See: Create Amazon Knowledge Base

  • To use additional models from Amazon Bedrock that are not available for selection when creating a Model connection, you would have to procure the Model ID and the Model ARN from the supported models of Amazon Bedrock.

    See: Add Amazon Bedrock models from AWS Services

Prerequisites

The Automation Admin requires these roles and permissions to create and manage Model connections for their business organization.
  • Role: AAE_Basic, Automation Admin custom role
  • Permission: Attended Bot Runner
  • Settings: AI Data Management must be enabled by the Automation Admin and the check box selected for Allow users to disable logs on AI Skills. Allow users with the Bot Creator license to disable data logging when using AI Skills to enable the Data logging toggle in the AI Skills screen.

See Roles and permissions for AI Tools for the Automation Admin custom role permissions.

Other requirements:

  • As mentioned earlier, you would first create an Amazon Knowledge Base to create a Grounded by knowledge base Model connection and use it successfully in an AI Skill.
  • If you want to store authentication details in a credential vault, have that information handy. See Secure credential store through Credential Vault.
  • To test a Model connection, you must be connected to a Bot Agent 22.60.10 and later. As part of the test, you would have to run the bot on your desktop. Hence ensure the Bot Agent is configured to your user. For this task, if you have to switch connection to a different Control Room, see: Switch device registration between Control Room instances.
  • You would need access to the Recorder package and the AI Skills package to test the connection successfully. A test Prompt would be executed to test the Model connection.

Procedure

  1. In your Control Room environment, navigate to AI > Model connections > Create model connection.
  2. In the Create model connection screen you would configure these Connection settings:
    Create Grounded Model connections with Amazon Bedrock RAG capability
    You can manually enter the model name in the Choose a model or create a custom one field. The name you enter will be used for creating the Model connection.
    1. Model connection name: Provide a name for easy identification of the Model connection.
    2. Description (optional): Add a meaningful short description defining the connection.
    3. Choose a vendor: Choose a foundational model vendor from the supported list of vendors. For creating a Grounded by knowledge base Model connection with Amazon Bedrock, you would select Amazon Bedrock from the drop-down list.
    4. Choose a type: Choose Grounded by knowledge base for using the RAG capability.
    5. Choose a model or create a custom one: Choose a model from the drop-down list of validated models from Amazon Bedrock.
      Additionally, we also support other models available from Amazon Bedrock, which are not available in the drop-down list. You can use either the Model ID or Model ARN to add to the list. See: Add Amazon Bedrock models from AWS Services.
    6. Click Next to proceed to the Authentication details section.
  3. In the Authentication details section, configure these settings:
    1. Region: Choose the region where your selected model is deployed for authenticating the Model connection. You can also add a region that is not available in the drop-down list by referring to the list in Amazon Bedrock. Enter in this format to get the region added to the list. For example: us-east-1.
      For a list of supported deployment regions for Amazon Bedrock models, see Supported regions and models for Amazon Bedrock knowledge bases.
    2. Knowledge base ID: Provide the Knowledge Base ID you procured from Amazon Bedrock. .
      See Amazon Bedrock knowledge bases
    3. Access Key: This AWS access key serves as your unique identifier within the AWS ecosystem. It is a fundamental part of the authentication process, allowing AWS Services to recognize and validate your access.
    4. Secret Access Key: This key is the confidential counterpart to your Access Key ID. This key is used to sign requests made to AWS, enhancing security by ensuring that only authorized individuals or systems can access your AWS resources.
    5. Session Token (optional): This is an optional field. In addition to the above information, you can include a Session token, which is a temporary, time-bound token used when working with temporary security credentials. It provides an added layer of security, particularly in scenarios where temporary access is required, such as when using temporary security credentials.
    6. After setting up the authentication details, confirm and click Next to proceed to the Test connection section to test the Model connection.
    Note: For details on setting up the Access Key, Secret Access Key, and Session Token for Amazon Bedrock, see: Amazon Bedrock: Authenticate action.
  4. Click Test connection to make sure all connection details have been defined correctly and check if the connection is working.
    This is a desktop operation using a Bot Agent. Use Bot Agent 22.60.10 and later for successful testing.
    • If the connection works as expected, the system will process the request and you will get a system generated success message.
    • If the connection does not work as expected, you will get a system-generated message stating the reason for the connection failure. For example, if you have not downloaded the supported foundational model package to your workspace, you would get an error message. You would have to download the package and then retest the Model connection.
    • If the testing a Model connection is unsuccessful or if you leave the task incomplete, the Model connection will not get saved and you will have to restart the process of creating the Model connection.
  5. Click Next to proceed to the Invite roles section to begin assigning custom roles to users.
    The Automation Admin would create custom roles and assign the Model connections to the role, which can then be assigned to users. Only users assigned to this custom role can use this Model connection.
  6. Assign access to the Pro Developer via custom role (using RBAC), for using this Model connection to create an AI Skill.
  7. Click Create model connection to complete creating the Model connection.
    After successfully creating the Model connection, the Pro Developer would use it to create an AI Skill

    See: Create AI Skills with Grounded by knowledge base Model connections

    .

Next steps

After creating and testing the Model connection, you would assign it to the Pro Developers, who would use this connection to create AI Skills.

See Create AI Skills with Grounded by knowledge base Model connections.

Note: When you create or test an AI Skill in the AI Skill screen, the success or failure details along with the model responses can be viewed in these navigation screens:
  • Administration > AI governance > AI Prompt log
  • Administration > AI governance > Event log
  • Administration > Audit log

See AI governance.

As the next step in your sequence of tasks, go to Create AI Skills with Grounded by knowledge base Model connections and create an AI Skill and connect to a Grounded by knowledge base Model connection to eventually use it in an automation.