Bedrock adapter

Uses the Amazon Bedrock Converse API with a named AWS profile. The profile supplies credentials and region through the AWS SDK configuration.

Provider section shape

Add the definition to your user provider catalog:

{
  "providers": {
    "my-bedrock": {
      "adapter": "bedrock",
      "enabled": true,
      "profile": "my-aws-profile",
      "timeout_seconds": 45,
      "models": [
        {
          "model": "us.anthropic.claude-opus-4-6-v1",
          "context_window_tokens": 1000000,
          "max_output_tokens": 4096
        }
      ]
    }
  }
}

This example defines a separate Bedrock provider. To customize the built-in bedrock entry, use that name, omit adapter, and include only the fields you want to change.

Provider fields

The defaults below are the values in the built-in bedrock definition. A new provider must declare these fields explicitly.

Field	Description	Default
enabled	Makes the provider available.	true
profile	Named AWS profile used by the SDK.	default
timeout_seconds	Positive read timeout in seconds for the Bedrock client.	45
models	Available models; replaces the complete inherited list.	From the included catalog

Credentials

Configure the AWS profile before using it. The profile must resolve credentials and a region, and its identity needs permission to invoke the selected model or inference profile.

Provider configuration does not create AWS credentials or grant model access. For guided setup, see Add an LLM provider.

Models

Field	Description	Default
model	Model or inference profile identifier supported by Converse in your AWS setup.	Required for each model entry
context_window_tokens	Positive context window size in tokens.	Required for each model entry
max_output_tokens	Positive output token limit, sent as inferenceConfig.maxTokens.	Not set; no explicit limit sent

Defining models replaces the full inherited list. Replace the example model and token limits with values available to your AWS account and region.

For agent-level provider, model, and effort selection, see Model selection.