Skip to main content

Amazon Bedrock

Maven Dependency​

<dependency>
    <groupId>dev.langchain4j</groupId>
    <artifactId>langchain4j-bedrock</artifactId>
    <version>1.0.1-beta6</version>
</dependency>

AWS credentials​

In order to use Amazon Bedrock models, you need to configure AWS credentials. One of the options is to set the AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables. More information can be found here.

Difference Between InvokeAPI and ConverseAPI​

Amazon Bedrock offers two primary model invocation API operations for inference:

  • Converse – Amazon recommend using the Converse API as it provides consistent API, that works with all Amazon Bedrock models that support messages.
  • InvokeModel – Originally aimed at single calls to obtain a response to a single prompt.

ChatModel using ConverseAPI​

Guardrails is not supported by the current implementation.

Supported models and their features can be found here.

Models ids can be found here.

Configuration​

ChatModel model = BedrockChatModel.builder()
        .modelId("us.amazon.nova-lite-v1:0")
        .region(...)
        .maxRetries(...)
        .timeout(...)
        .logRequests(...)
        .logResponses(...)
        .listeners(...)
        .defaultRequestParameters(BedrockChatRequestParameters.builder()
                .topP(...)
                .temperature(...)
                .maxOutputTokens(...)
                .stopSequences(...)
                .toolSpecifications(...)
                .additionalModelRequestFields(...)
                .build())
        .build();

The field additionalModelRequestFields is a Map<String, Object>. As explained here it allows to add inference parameters for a specific model that is not covered by common inferenceConfig. BedrockChatRequestParameters has a convenience method to enable Claude 3.7 thinking process through adding inference parameters in additionalModelRequestFields.

Examples​

StreamingChatModel using ConverseAPI​

Guardrails is not supported by the current implementation.

Supported models and their features can be found here.

Models ids can be found here.

Configuration​

StreamingChatModel model = BedrockStreamingChatModel.builder()
        .modelId("us.amazon.nova-lite-v1:0")
        .region(...)
        .maxRetries(...)
        .timeout(...)
        .logRequests(...)
        .logResponses(...)
        .listeners(...)
        .defaultRequestParameters(BedrockChatRequestParameters.builder()
                .topP(...)
                .temperature(...)
                .maxOutputTokens(...)
                .stopSequences(...)
                .toolSpecifications(...)
                .additionalModelRequestFields(...)
                .build())
        .build();

The field additionalModelRequestFields is a Map<String, Object>. As explained here it allows to add inference parameters for a specific model that is not covered by common inferenceConfig. BedrockChatRequestParameters has a convenience method to enable Claude 3.7 thinking process through adding inference parameters in additionalModelRequestFields.

Examples​

ChatModel using InvokeAPI​

AI21 Models​

  • BedrockAI21LabsChatModel (deprecated, please use BedrockChatModel)

Anthropic Models​

  • BedrockAnthropicMessageChatModel: (deprecated, please use BedrockChatModel) supports new Messages API
  • BedrockAnthropicCompletionChatModel: (deprecated, please use BedrockChatModel) supports old Text Completions API
  • BedrockAnthropicStreamingChatModel

Example:

ChatModel model = BedrockAnthropicMessageChatModel.builder()
.model("anthropic.claude-3-sonnet-20240229-v1:0")
.build();

Cohere Models​

  • BedrockCohereChatModel (deprecated, please use BedrockChatModel)

Meta Llama Models​

  • BedrockLlamaChatModel (deprecated, please use BedrockChatModel)

Mistral Models​

  • BedrockMistralAiChatModel (deprecated, please use BedrockChatModel)

Titan Models​

  • BedrockTitanChatModel (deprecated, please use BedrockChatModel)
  • BedrockTitanEmbeddingModel

Examples​