Skip to main content
AWS Bedrock is MIRA’s default provider for the RLM Engine. It gives you access to Anthropic’s Claude models through your existing AWS account without a direct Anthropic API subscription. Both the RLM and Native Agent engines support Bedrock.

​
Prerequisites

  • An AWS account with Bedrock enabled in your target region
  • IAM access keys (or an IAM role if running on EC2/ECS)
  • Model access granted in the Bedrock console for the model you want to use

​
Request model access

Before you can invoke a model, you must request access in the AWS console:
  1. Open the AWS Bedrock console
  2. Go to Model access in the left sidebar
  3. Find the model (e.g. Anthropic Claude Sonnet 4) and click Request access
  4. Access is typically granted within a few minutes

​
Setup in MIRA

1

Enter your AWS credentials

Press ⌘, → Bedrock / AWS tab and fill in:
  • AWS Access Key ID — from IAM → Users → Security credentials
  • AWS Secret Access Key — shown once when you create the access key
  • Session Token — optional, only for temporary/assumed-role credentials
  • AWS Region — where you have Bedrock enabled (e.g. eu-west-1)
Click Save. Credentials are encrypted with safeStorage and stored in credentials.json.
2

Test the connection

Click Test Connection. MIRA calls sts:GetCallerIdentity and displays your AWS Account ID on success.
3

Select Bedrock in the Engine tab

Press ⌘, → Engine tab → in the Provider & Model section, click the AWS Bedrock button.
4

Enter a model ID and region

MIRA shows suggestion chips for common Bedrock model IDs. You can also type any valid Bedrock model ID directly — any model you have access to in your account works.Select the AWS Region that matches your Bedrock model access.
5

Save

Click Save & restart bridge. The engine restarts automatically.
Bedrock supports many foundation models beyond Claude (e.g. Llama, Mistral, Amazon Titan). Paste any valid Bedrock model ID from the AWS Bedrock docs into the Model ID field. Note that the RLM engine is tested primarily with Claude and GPT-class models.

​
Required IAM permissions

The minimum IAM policy required to run MIRA with Bedrock:

​
Cross-region inference

The eu. and us. prefixed model IDs use Bedrock’s cross-region inference feature, which routes requests across availability zones within that geographic area for higher availability. You do not need to enable anything extra — just ensure the region in your settings matches the prefix (eu-west-1 for eu. models, us-east-1 or similar for us. models).

​
Default values

MIRA uses these Bedrock defaults (configurable in the Engine tab):

​
Troubleshooting

Edit this page — Open a pull request