bedrock_nova

AWS Bedrock Nova Example

This example demonstrates how to use Amazon Nova models through AWS Bedrock with LangGraph-Go.

Overview

Amazon Nova is AWS's family of foundation models, available through AWS Bedrock. This example shows:

1. Basic Chat - Simple question-answer interaction
2. Tool Calling - Using Nova's function calling capabilities
3. Multi-Turn Conversations - Maintaining context across turns
4. Model Variants - Comparing Nova Micro, Lite, and Pro

Prerequisites

1. AWS Account Setup

You need an AWS account with:

• AWS Bedrock access enabled
• Model access granted for Nova models in your region

2. Enable Model Access

1. Go to AWS Console → Amazon Bedrock → Model access
2. Request access for:
   • Amazon Nova Micro
   • Amazon Nova Lite
   • Amazon Nova Pro
3. Wait for approval (usually instant for Nova models)

3. AWS Credentials

Configure AWS credentials using one of these methods:

Environment Variables:

export AWS_ACCESS_KEY_ID=your_access_key
export AWS_SECRET_ACCESS_KEY=your_secret_key
export AWS_REGION=us-east-1

AWS Credentials File (~/.aws/credentials):

[default]
aws_access_key_id = your_access_key
aws_secret_access_key = your_secret_key
region = us-east-1

IAM Role (for EC2/ECS/Lambda): No configuration needed - automatically uses instance role.

Running the Example

# From project root
cd examples/bedrock_nova

# Run the example
go run main.go

Expected Output

=== Example 1: Basic Chat ===
Question: What is the capital of France?
Nova Lite: The capital of France is Paris.
Tokens: 12 input, 8 output

=== Example 2: Tool Calling ===
Question: What's the weather like in Tokyo?
Nova wants to call tool: get_weather
With arguments: map[location:Tokyo]

=== Example 3: Multi-Turn Conversation ===
User: My favorite color is blue.
Nova: That's great! Blue is a wonderful color...

User: What color did I just mention?
Nova: You mentioned that your favorite color is blue.

=== Example 4: Nova Model Variants ===

--- Nova Micro (Fastest, lowest cost) ---
Response: 4
Tokens: 8 in, 1 out

--- Nova Lite (Balanced speed and capability) ---
Response: The answer is 4.
Tokens: 8 in, 5 out

--- Nova Pro (Most capable, multimodal) ---
Response: 2 + 2 equals 4.
Tokens: 8 in, 7 out

Nova Model Variants

Configuration Options

config := bedrock.Config{
    Region:          "us-east-1",              // AWS region
    ModelID:         "amazon.nova-lite-v1:0",  // Model ID
    MaxTokens:       512,                       // Max response tokens
    Temperature:     0.7,                       // 0.0-1.0 (creativity)
    TopP:            0.9,                       // Nucleus sampling
    StopSequences:   []string{"\n\n"},         // Stop generation
    FallbackRegions: []string{"us-west-2"},    // Regional failover
}

Inference Profiles

Use inference profiles for automatic cross-region routing:

// Inference profile (recommended for production)
ModelID: "us.amazon.nova-lite-v1:0"

// Direct model ID (single region)
ModelID: "amazon.nova-lite-v1:0"

Inference profiles provide:

• Automatic regional failover
• Better availability
• Load balancing across regions

Error Handling

response, err := adapter.Chat(ctx, messages, nil)
if err != nil {
    // Check for specific error types
    var bedrockErr *bedrock.BedrockError
    if errors.As(err, &bedrockErr) {
        switch bedrockErr.Code {
        case "ModelNotReadyException":
            // Model loading
        case "ThrottlingException":
            // Rate limited
        case "ValidationException":
            // Invalid request
        }
    }
}

Troubleshooting

"Model access denied"

• Go to AWS Console → Bedrock → Model access
• Request access for Nova models in your region

"Region not supported"

• Nova models are available in: us-east-1, us-west-2, eu-west-1, ap-southeast-1
• Check AWS documentation for latest regional availability

"AWS credentials not found"

• Verify AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY are set
• Check ~/.aws/credentials file exists and is readable
• Ensure IAM role has bedrock:InvokeModel permission

"Throttling errors"

• Implement exponential backoff (built into adapter)
• Request quota increase in AWS Service Quotas console
• Use inference profiles for better rate limits

Next Steps

• Explore streaming responses with ChatStream()
• Implement tool execution loops
• Add conversation memory with checkpointing
• Use multi-region failover for production resilience

References

• AWS Bedrock Documentation
• Amazon Nova Models
• LangGraph-Go Bedrock Adapter