Several Model Families Inside Your Own AWS Account

What you can do together

• Reach several model families through one provider, so changing model is a configuration change inside an existing AWS relationship.
• Keep inference in the AWS region and account your commitments require, including GovCloud.
• Scope what the platform may invoke through the IAM role rather than granting broadly.
• Run Bedrock for one agent and a self-hosted model for another, where material differs in sensitivity.

How it connects

The model behind an agent or a node is configuration. Bedrock is selected per deployment or per agent, and both the Bedrock and the Bedrock Converse interfaces are carried.

Inference runs in the AWS region and account you point it at, under the IAM role the connection is configured with. That role is the boundary: what the platform can invoke is what the role is granted, and scoping it is the control rather than relying on who is asking.

GovCloud is the case worth naming. Where a federal or state buyer requires an AWS GovCloud boundary, Bedrock model availability in that partition is narrower than in the commercial regions and differs by model family, so an engagement confirms which models are actually available in the target region before a workflow is designed around one. That is an AWS fact rather than ours, and it changes designs late if it is discovered late.