Amazon Bedrock

AWS's fully managed multi-agent framework. Supervisor/collaborator model maps directly to the Envelope team hierarchy.

Overview

Amazon Bedrock Agents is a fully managed service — AWS handles agent infrastructure, model routing, and invocation. You bring an Envelope team definition; Envelope provisions Bedrock Agents via the AWS SDK, wires up the supervisor/collaborator relationships, and returns provisioned agent ARNs.

Bedrock supports all major LLM providers through its model library — Anthropic Claude, Amazon Titan, Meta Llama, Mistral, and others. Each agent's modelConfig in the Envelope schema maps to a Bedrock Foundation Model ARN at provisioning time.

Concept mapping

| Envelope concept | Bedrock equivalent | |---|---| | Team | Agent group / supervisor + collaborators | | Agent | Bedrock Agent | | Run | InvokeAgent API call | | Install | Agent + alias provisioning | | reportsToKey | Supervisor routing via collaborator agents | | requiredSecrets | Agent action group secrets |

IAM setup

Before deploying, create an IAM role for Bedrock agents to assume.

Trust policy

{
  "Principal": { "Service": "bedrock.amazonaws.com" },
  "Action": "sts:AssumeRole"
}

Required permissions

bedrock:InvokeModel
bedrock:InvokeAgent
bedrock:CreateAgent
bedrock:CreateAgentAlias

Supply the ARN of this role as agentResourceRoleArn at install time. All provisioned agents assume this role when invoked.

Credentials

| Key | Description | |---|---| | accessKeyId | AWS Access Key ID — from your IAM user's security credentials | | secretAccessKey | AWS Secret Access Key — paired with the Access Key ID | | region | AWS region where Bedrock is enabled (e.g. us-east-1, eu-west-1) | | agentResourceRoleArn | ARN of the IAM role each Bedrock agent assumes at runtime — must trust bedrock.amazonaws.com with bedrock:InvokeModel permission |

Deploying a team

Use the guided deploy wizard to walk through platform selection, IAM credential entry, and variable configuration. Envelope validates the credentials and provisions agents in your AWS account before returning a dashboard URL.

Region availability — Bedrock model availability varies by region. Verify that your target models (e.g. Claude on Anthropic) are available in your chosen region before deploying. US East (us-east-1) has the broadest model coverage.