Extract data using Anthropic models

You can use the Anthropic generative AI models that are available via AWS and GCP for data extraction in Document Automation. The Anthropic option is available from v.33 and later releases.

Prerequisites

By default, no additional configuration is required for Anthropic. Data extraction will work using an Automation Anywhere service account available in AWS. However, if you want to use your own account, ensure that you have performed the following tasks depending on the Cloud provider you have configured for Anthropic:

Anthropic provides the following advantages:

  • Efficient processing of large, unstructured documents
  • Handle documents in both English and other languages
  • Faster processing of documents with better data extraction accuracy
  • Flexibility to use your own license via the bring your own license (BYOL) model
Note: We recommend that you use Anthropic Claude 3.0 or later models for data extraction.

Procedure

  1. Navigate to Automation > Document Workspace Processes.
  2. Click the folder with the same name as the learning instance. For example, if the learning instance name is Residential Lease, then the folder name will be Residential Lease.
  3. Click <li_name>_extractionbot.
  4. In the Bot editor, select the Extract data action.
  5. In the Additional settings option, select Anthropic.
  6. Use one of the following options depending on the Cloud provider you have configured for Anthropic:
    • AWS Bedrock
      1. In the Access key option, enter the unique identifier associated with a user that is required for authentication.

        The access key is used as the username for authentication. Use the Credential, Variable, or Insecure string option to enter the access key.

      2. In the Secret access key option, enter the secret string associated with the access key that is required for authentication.

        The secret access key is used as the password for authentication. Use the Credential, Variable, or Insecure string option to enter the secret access key.

      3. (Optional) In the Session token option, enter a short-lived security credential that provides temporary access to the service.

        This is an optional configuration and is required only if you want users to have access for a limited period. Use the Credential, Variable, or Insecure string option to enter the session token.

        Note: If you are using the session token option, ensure that you update this token whenever it is refreshed to process documents without interruptions or errors.
      4. In the Endpoint URL for Claude model option, enter the URL to specify the Anthropic model and to send requests to the AWS Bedrock endpoint.

        For example, https://bedrock-runtime.{aws-region}.amazonaws.com/model/{model-id}/invoke. See Amazon Bedrock endpoints and quotas and AWS Bedrock model IDs .

    • GCP Vertex AI
      1. In the Service account key option, enter the credentials used to authenticate with GCP services.

        Use the Credential, Variable, or Insecure string option to enter the service account key. See Creating a service account.

        Note: Google refreshes the private key value in the service account key at certain intervals for security reasons. Ensure that you update this value whenever the private key value is refreshed to process documents without interruptions or errors.
      2. In the Endpoint URL for Claude model option, enter the URL to specify the Anthropic model and to send requests to the Google Vertex AI endpoint.

        For example, https://{gcp-region}-aiplatform.googleapis.com/v1/projects/{project-id}/locations/{deployment-region}/publishers/anthropic/models/{model-ID}:RawPredict. See AWS Bedrock InvokeModel examples.

    Note: Verify that the Anthropic settings are correctly configured. Otherwise, an error is displayed when documents are processed in the learning instances.
Now that you have configured BYOL to use the Anthropic model, you can process documents using the learning instance to extract data.