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

Guardrails

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

AWS Bedrock Guardrails provide content filtering, safety checks, and policy enforcement.

Capabilities

Filter harmful content: Block toxic, violent, sexual, or hateful content
Detect PII: Identify and anonymize personally identifiable information
Block topics: Prevent discussions of specific unwanted topics
Filter words: Block profanity and custom word lists
Ensure grounding: Verify responses are grounded in provided context

How It Works

Create guardrail in AWS Console with policies
Reference guardrail by ID and version in requests
Receive assessments in response metadata

AWS Bedrock Guardrails Documentation

Quick Example

import dev.langchain4j.model.bedrock.BedrockGuardrailConfiguration;
import dev.langchain4j.model.bedrock.BedrockChatRequestParameters;
import dev.langchain4j.model.bedrock.BedrockChatModel;

// Configure guardrail
BedrockGuardrailConfiguration guardrail = BedrockGuardrailConfiguration.builder()
    .guardrailIdentifier("my-guardrail-id")
    .guardrailVersion("1")
    .build();

BedrockChatRequestParameters params = BedrockChatRequestParameters.builder()
    .guardrailConfiguration(guardrail)
    .build();

BedrockChatModel model = BedrockChatModel.builder()
    .modelId("anthropic.claude-3-5-sonnet-20241022-v2:0")
    .defaultRequestParameters(params)
    .build();

// Chat with guardrail protection
ChatResponse response = model.chat(request);

// Check for violations
if (response.metadata() instanceof BedrockChatResponseMetadata metadata) {
    GuardrailAssessmentSummary summary = metadata.guardrailAssessmentSummary();
    if (summary != null) {
        handleAssessments(summary);
    }
}

Policy Types

public enum Policy {
    TOPIC,      // Blocks unwanted topics
    CONTENT,    // Filters harmful content (hate, violence, sexual, etc.)
    WORD,       // Blocks profanity and custom word lists
    SENSITIVE,  // Detects and anonymizes PII
    CONTEXT     // Verifies factual accuracy
}

Actions

public enum Action {
    ANONYMIZED,  // PII was masked/removed
    BLOCKED,     // Content was blocked
    NONE,        // No action (passed policy check)
    UNKNOWN      // Unrecognized action
}

API Classes

BedrockGuardrailConfiguration

public class BedrockGuardrailConfiguration {
    public BedrockGuardrailConfiguration(String guardrailIdentifier, String guardrailVersion);

    public String guardrailIdentifier();
    public String guardrailVersion();

    public static Builder builder();
}

GuardrailAssessmentSummary

public class GuardrailAssessmentSummary {
    public List<GuardrailAssessment> inputAssessments();
    public List<GuardrailAssessment> ouputAssessments();

    public static Builder builder();
}

GuardrailAssessment

public class GuardrailAssessment {
    public Action action();
    public Policy policy();
    public String name();

    public static Builder<?> builder();
}

IAM Permissions

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream",
        "bedrock:ApplyGuardrail"
      ],
      "Resource": [
        "arn:aws:bedrock:*::foundation-model/*",
        "arn:aws:bedrock:*:*:guardrail/*"
      ]
    }
  ]
}

Next Steps

Configuration Guide - Set up guardrails
Handling Assessments - Process violations
Use Cases - Common scenarios

Install with Tessl CLI

npx tessl i tessl/maven-dev-langchain4j--langchain4j-bedrock@1.11.0