tessl/maven-dev-langchain4j--langchain4j-bedrock

AWS Bedrock integration for LangChain4j enabling Java applications to interact with various LLM providers through a unified interface

Overview

Eval results

Files

Guardrail Configuration

Name: tessl/maven-dev-langchain4j--langchain4j-bedrock
Author: tessl

How to configure and apply guardrails to your requests.

Creating a Guardrail Configuration

import dev.langchain4j.model.bedrock.BedrockGuardrailConfiguration;

// With ID and version
BedrockGuardrailConfiguration guardrail = BedrockGuardrailConfiguration.builder()
    .guardrailIdentifier("abc123xyz")
    .guardrailVersion("1")
    .build();

// Or using constructor
BedrockGuardrailConfiguration guardrail = new BedrockGuardrailConfiguration(
    "abc123xyz",  // Guardrail ID
    "1"           // Version
);

Version Management

// Use specific version in production
BedrockGuardrailConfiguration production = BedrockGuardrailConfiguration.builder()
    .guardrailIdentifier("my-guardrail")
    .guardrailVersion("1")  // Stable version
    .build();

// Use DRAFT for development/testing
BedrockGuardrailConfiguration development = BedrockGuardrailConfiguration.builder()
    .guardrailIdentifier("my-guardrail")
    .guardrailVersion("DRAFT")  // Latest changes
    .build();

Using ARN Instead of ID

// With full ARN
BedrockGuardrailConfiguration arnConfig = BedrockGuardrailConfiguration.builder()
    .guardrailIdentifier("arn:aws:bedrock:us-east-1:123456789012:guardrail/abc123xyz")
    .guardrailVersion("1")
    .build();

Applying to Request Parameters

import dev.langchain4j.model.bedrock.BedrockChatRequestParameters;

// Include in request parameters
BedrockChatRequestParameters params = BedrockChatRequestParameters.builder()
    .guardrailConfiguration(guardrail)
    .temperature(0.7)
    .maxOutputTokens(2048)
    .build();

Applying to Model (Default for All Requests)

import dev.langchain4j.model.bedrock.BedrockChatModel;

BedrockChatModel model = BedrockChatModel.builder()
    .region(Region.US_EAST_1)
    .modelId("anthropic.claude-3-5-sonnet-20241022-v2:0")
    .defaultRequestParameters(
        BedrockChatRequestParameters.builder()
            .guardrailConfiguration(guardrail)
            .build()
    )
    .build();

// All requests use this guardrail
ChatResponse response = model.chat(request);

Per-Request Override

// Model with default guardrail
BedrockChatModel model = BedrockChatModel.builder()
    .modelId("anthropic.claude-3-5-sonnet-20241022-v2:0")
    .defaultRequestParameters(
        BedrockChatRequestParameters.builder()
            .guardrailConfiguration(defaultGuardrail)
            .build()
    )
    .build();

// Override with stricter guardrail for specific request
BedrockChatRequestParameters strictParams = BedrockChatRequestParameters.builder()
    .guardrailConfiguration(strictGuardrail)
    .build();

ChatResponse response = model.chat(ChatRequest.builder()
    .messages(UserMessage.from("User input"))
    .parameters(strictParams)  // Override default
    .build());

Configuration Reuse

// Reuse guardrail configuration
private static final BedrockGuardrailConfiguration CONTENT_FILTER =
    BedrockGuardrailConfiguration.builder()
        .guardrailIdentifier("content-filter")
        .guardrailVersion("1")
        .build();

private static final BedrockGuardrailConfiguration PII_DETECTOR =
    BedrockGuardrailConfiguration.builder()
        .guardrailIdentifier("pii-detector")
        .guardrailVersion("1")
        .build();

// Use in different contexts
BedrockChatRequestParameters publicParams = BedrockChatRequestParameters.builder()
    .guardrailConfiguration(CONTENT_FILTER)
    .build();

BedrockChatRequestParameters internalParams = BedrockChatRequestParameters.builder()
    .guardrailConfiguration(PII_DETECTOR)
    .build();

Creating Guardrails in AWS

Guardrails must be created in AWS Bedrock Console:

Navigate to AWS Bedrock Console → Guardrails
Create guardrail and configure policies:
- Content filters: Set thresholds (NONE, LOW, MEDIUM, HIGH) for hate, violence, sexual content, etc.
- Denied topics: Define topics to block
- Word filters: Add profanity or custom word lists
- Sensitive information: Configure PII detection and anonymization
- Contextual grounding: Enable grounding checks
Version and publish the guardrail
Note the guardrail ID for use in code

Region Considerations

Guardrails must be in the same region as the model
Reference guardrail ARN to ensure correct region

// Ensure guardrail and model in same region
BedrockChatModel model = BedrockChatModel.builder()
    .region(Region.US_EAST_1)  // Model in us-east-1
    .modelId("anthropic.claude-3-5-sonnet-20241022-v2:0")
    .defaultRequestParameters(
        BedrockChatRequestParameters.builder()
            .guardrailConfiguration(
                BedrockGuardrailConfiguration.builder()
                    .guardrailIdentifier("arn:aws:bedrock:us-east-1:123456789012:guardrail/xyz")  // Same region
                    .guardrailVersion("1")
                    .build()
            )
            .build()
    )
    .build();

Next: Handling Assessments to process guardrail violations

Install with Tessl CLI

npx tessl i tessl/maven-dev-langchain4j--langchain4j-bedrock