Connect your own generative AI services
- Updated: 2024/12/03
Connect your own generative AI services
If your organization’s security protocols or data privacy policies require processing documents within your own infrastructure and prohibit sharing data or documents with external vendors (except for providers such as Amazon, Google, or Microsoft), you will need to connect your own generative AI service account to Document Automation.
Prerequisites
- Anthropic
- Anthropic models are supported when provided through Amazon Bedrock and GCP Vertex AI.
Native Anthropic services are not supported.
- AWS Bedrock
- Set up Amazon Bedrock. See Set up Amazon Bedrock.
- Added Anthropic Claude model access. See Add or remove access to Amazon Bedrock foundation models.
- Configured appropriate model access permissions. See Grant IAM permissions to request access to Amazon Bedrock foundation models.
- Made note of your access key and secret access keys. See Manage access keys for IAM users.
- GCP Vertex AI
- Configured the required roles and permissions to use the Anthropic Claude models. See Use the Claude 3 models from Anthropic.
- Created a service account. See Creating a service account.
- Created a service account key (.json format). See Creating a service account.
- AWS Bedrock
- Azure OpenAI
-
- Ensure that you have created and deployed an Azure OpenAI Service resource. See Create and deploy an Azure OpenAI Service resource.
- Ensure that you have made a note of your access key (KEY 1 or KEY 2) and endpoint for GPT and embeddings. See See Azure OpenAI GPT-4o-mini fine-tuning tutorial and Tutorial: Explore Azure OpenAI Service embeddings and document search.
When creating a learning instance, you can select a generative AI service, such as Azure OpenAI or Anthropic. By default, Document Automation uses the Automation Anywhere service account, but if you need greater control over data flow, you can connect your own credentials within the extraction bot of the learning instance. While the same credentials can be used for multiple learning instances, each instance must be explicitly configured. The exact procedure for this is detailed below.
To securely store and share credentials with other users or Bot Runners, you must create and save your credentials, store your credentials in a locker, and select the users who can use the credentials.