Amazon Bedrock

Amazon Bedrock is a fully managed service on AWS that provides access to foundation models from various AI companies through a single API.

Chat model

We recommend configuring Claude 3.7 Sonnet as your chat model.

YAML

config.yaml

models:
  - name: Claude 3.7 Sonnet
    provider: bedrock
    model: us.anthropic.claude-3-7-sonnet-20250219-v1:0
    env:
      region: us-east-1
      profile: bedrock
    roles:
      - chat

JSON

config.json

{
  "models": [
    {
      "title": "Claude 3.7 Sonnet",
      "provider": "bedrock",
      "model": "us.anthropic.claude-3-7-sonnet-20250219-v1:0",
      "region": "us-east-1",
      "profile": "bedrock"
    }
  ]
}

If you run into the following error when connecting to the new Claude 3.5 Sonnet 2 models from AWS - 400 Invocation of model ID anthropic.claude-3-5-sonnet-20241022-v2:0 with on-demand throughput isn't supported. Retry your request with the ID or ARN of an inference profile that contains this model.

You can fix this using the following config:

YAML

config.yaml

models:
  - name: Claude 3.5 Sonnet
    provider: bedrock
    model: us.anthropic.claude-3-5-sonnet-20241022-v2:0
    env:
      region: us-east-1
      profile: bedrock
    roles:
      - chat

JSON

config.json

{
  "models": [
    {
      "title": "Claude 3.5 Sonnet",
      "provider": "bedrock",
      "model": "us.anthropic.claude-3-5-sonnet-20241022-v2:0",
      "region": "us-east-1",
      "profile": "bedrock"
    }
  ]
}

Autocomplete model

Bedrock currently does not offer any autocomplete models. However, Codestral from Mistral and Point from Poolside will be supported in the near future.

In the meantime, you can view a list of autocomplete model providers here.

Embeddings model

We recommend configuring amazon.titan-embed-text-v2:0 as your embeddings model.

YAML

config.yaml

models:
  - name: Embeddings Model
    provider: bedrock
    model: amazon.titan-embed-text-v2:0
    env:
      region: us-west-2
    roles:
      - embed

JSON

config.json

{
  "embeddingsProvider": {
    "title": "Embeddings Model",
    "provider": "bedrock",
    "model": "amazon.titan-embed-text-v2:0",
    "region": "us-west-2"
  }
}

Reranking model

We recommend configuring cohere.rerank-v3-5:0 as your reranking model, you may also use amazon.rerank-v1:0.

YAML

config.yaml

models:
  - name: Bedrock Reranker
    provider: bedrock
    model: cohere.rerank-v3-5:0
    env:
      region: us-west-2
    roles:
      - rerank

JSON

config.json

{
  "reranker": {
    "name": "bedrock",
    "params": {
      "model": "cohere.rerank-v3-5:0",
      "region": "us-west-2"
    }
  }
}

Prompt caching

Bedrock allows Claude models to cache tool payloads, system messages, and chat messages between requests. Enable this behavior by adding promptCaching: true under defaultCompletionOptions in your model configuration.

Prompt caching is generally available for:

• Claude 3.7 Sonnet
• Claude 3.5 Haiku
• Amazon Nova Micro
• Amazon Nova Lite
• Amazon Nova Pro

Customers who were granted access to Claude 3.5 Sonnet v2 during the prompt caching preview will retain that access, but it cannot be enabled for new users on that model.

config.yaml

models:
  - name: Claude 3.7 Sonnet
    provider: bedrock
    model: us.anthropic.claude-3-7-sonnet-20250219-v1:0
    defaultCompletionOptions:
      promptCaching: true

Prompt caching is not supported in JSON configuration files, so use the YAML syntax above to enable it.

Authentication

Authentication will be through temporary or long-term credentials in ~/.aws/credentials under a configured profile (e.g. "bedrock").

[bedrock]
aws_access_key_id = abcdefg
aws_secret_access_key = hijklmno
aws_session_token = pqrstuvwxyz # Optional: means short term creds.

Custom Imported Models

To setup Bedrock using custom imported models, add the following to your config file:

YAML

config.yaml

models:
  - name: AWS Bedrock deepseek-coder-6.7b-instruct
    provider: bedrockimport
    model: deepseek-coder-6.7b-instruct
    modelArn: arn:aws:bedrock:us-west-2:XXXXX:imported-model/XXXXXX
    region: us-west-2
    profile: bedrock

JSON

config.json

{
  "models": [
    {
      "title": "AWS Bedrock deepseek-coder-6.7b-instruct",
      "provider": "bedrockimport",
      "model": "deepseek-coder-6.7b-instruct",
      "modelArn": "arn:aws:bedrock:us-west-2:XXXXX:imported-model/XXXXXX", 
      "region": "us-west-2",
      "profile": "bedrock"
    }
  ]
}

Authentication will be through temporary or long-term credentials in ~/.aws/credentials under a configured profile (e.g. "bedrock").

[bedrock]
aws_access_key_id = abcdefg
aws_secret_access_key = hijklmno
aws_session_token = pqrstuvwxyz # Optional: means short term creds.