Skip to main content

​
Overview

AWS Bedrock provides access to multiple foundation models from Amazon, Anthropic, Meta, Mistral, and others through a single API. Enterprise-grade with AWS security and compliance. Model Class: BedrockConverseModel

​
Authentication

​
AWS Credentials

Configure AWS credentials using one of these methods: 
# Environment variables
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."
export AWS_REGION="us-east-1"

# Or use AWS CLI
aws configure

​
Using infer_model

from upsonic import infer_model

# Format: bedrock/provider.model-id
model = infer_model("bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0")

​
Manual Configuration

from upsonic.models.bedrock import BedrockConverseModel, BedrockModelSettings

settings = BedrockModelSettings(
    max_tokens=2048,
    temperature=0.7
)

model = BedrockConverseModel(
    model_name="anthropic.claude-3-5-sonnet-20241022-v2:0",
    settings=settings
)

​
With Custom Region

from upsonic.providers.bedrock import BedrockProvider
import boto3

# Custom region and credentials
session = boto3.Session(
    region_name="us-west-2",
    aws_access_key_id="...",
    aws_secret_access_key="..."
)
client = session.client("bedrock-runtime")

provider = BedrockProvider(client=client)
model = BedrockConverseModel(
    model_name="anthropic.claude-3-5-sonnet-20241022-v2:0",
    provider=provider
)

​
Examples

​
Basic Usage

from upsonic import Agent, Task, infer_model

model = infer_model("bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0")
agent = Agent(model=model)

task = Task("Analyze this business report")
result = agent.do(task)

​
With Guardrails

from upsonic.models.bedrock import BedrockConverseModel, BedrockModelSettings

settings = BedrockModelSettings(
    max_tokens=2048,
    temperature=0.7,
    bedrock_guardrail_config={
        "guardrailIdentifier": "your-guardrail-id",
        "guardrailVersion": "1"
    }
)

model = BedrockConverseModel(
    model_name="anthropic.claude-3-5-sonnet-20241022-v2:0",
    settings=settings
)

​
With Request Metadata

from upsonic.models.bedrock import BedrockConverseModel, BedrockModelSettings

settings = BedrockModelSettings(
    max_tokens=2048,
    bedrock_request_metadata={
        "user_id": "user-123",
        "session_id": "session-456"
    }
)

model = BedrockConverseModel(
    model_name="anthropic.claude-3-5-sonnet-20241022-v2:0",
    settings=settings
)

​
With Amazon Nova

from upsonic import Agent, Task, infer_model

# Amazon's own models
model = infer_model("bedrock/us.amazon.nova-pro-v1:0")
agent = Agent(model=model)

task = Task("Generate a product description")
result = agent.do(task)

​
Cross-Region Inference

from upsonic import infer_model

# Use cross-region inference for better availability
model = infer_model("bedrock/us.anthropic.claude-3-5-sonnet-20241022-v2:0")

​
Prompt Caching

Bedrock supports prompt caching for supported models (primarily Claude): 
from upsonic import Agent, Task, infer_model

# Long context will be cached for Claude models
system_prompt = """
You are an AWS expert assistant with knowledge of:
- All AWS services and their use cases
- Best practices for cloud architecture
- Security and compliance requirements
... (extensive AWS documentation)
"""

model = infer_model("bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0")
agent = Agent(
    model=model,
    system_prompt=system_prompt
)

# Subsequent requests benefit from caching
task1 = Task("How do I set up S3 encryption?")
result1 = agent.do(task1)

task2 = Task("What's the best EC2 instance for my workload?")
result2 = agent.do(task2)

​
Cache Benefits

  • Cost Savings: Reduced input token costs
  • Latency: Faster processing
  • Automatic: No configuration for Claude models
  • Model-Specific: Check model documentation for support

​
Model Parameters

​
Base Settings

ParameterTypeDescriptionDefault
max_tokensintMaximum tokens to generate2048
temperaturefloatSampling temperature1.0
top_pfloatNucleus sampling1.0
stop_sequenceslist[str]Stop sequencesNone

​
Bedrock-Specific Settings

ParameterTypeDescription
bedrock_guardrail_configdictContent moderation settings
bedrock_performance_configurationdictPerformance optimization
bedrock_request_metadatadict[str, str]Request metadata for tracking
bedrock_additional_model_response_fields_pathslist[str]Additional response fields
bedrock_prompt_variablesdictTemplate variables
bedrock_additional_model_requests_fieldsdictModel-specific parameters

​
Guardrail Configuration

bedrock_guardrail_config = {
    "guardrailIdentifier": "abc123xyz",
    "guardrailVersion": "1",
    "trace": "enabled"  # Optional: enable tracing
}

​
Performance Configuration

bedrock_performance_configuration = {
    "latency": "optimized"  # or "standard"
}

​
Example Configuration

from upsonic.models.bedrock import BedrockConverseModel, BedrockModelSettings

settings = BedrockModelSettings(
    # Base settings
    max_tokens=4096,
    temperature=0.7,
    top_p=0.9,
    stop_sequences=["END"],
    
    # Bedrock-specific
    bedrock_guardrail_config={
        "guardrailIdentifier": "guardrail-id",
        "guardrailVersion": "1"
    },
    bedrock_performance_configuration={
        "latency": "optimized"
    },
    bedrock_request_metadata={
        "user_id": "user-123",
        "application": "chatbot"
    }
)

model = BedrockConverseModel(
    model_name="anthropic.claude-3-5-sonnet-20241022-v2:0",
    settings=settings
)

​
Available Models

​
Amazon Nova

  • us.amazon.nova-pro-v1:0: Balanced performance
  • us.amazon.nova-lite-v1:0: Fast and affordable
  • us.amazon.nova-micro-v1:0: Ultra-lightweight

​
Anthropic Claude

  • anthropic.claude-3-5-sonnet-20241022-v2:0: Latest Claude
  • us.anthropic.claude-3-5-sonnet-20241022-v2:0: Cross-region
  • anthropic.claude-3-5-haiku-20241022-v1:0: Fast variant
  • us.anthropic.claude-3-7-sonnet-20250219-v1:0: Enhanced capabilities
  • us.anthropic.claude-opus-4-20250514-v1:0: Most powerful

​
Meta Llama

  • meta.llama3-1-70b-instruct-v1:0: Large model
  • meta.llama3-1-8b-instruct-v1:0: Efficient model
  • us.meta.llama3-2-90b-instruct-v1:0: Latest large
  • us.meta.llama3-3-70b-instruct-v1:0: Latest generation

​
Mistral

  • mistral.mistral-large-2407-v1:0: Most capable
  • mistral.mixtral-8x7b-instruct-v0:1: Fast and efficient

​
Cohere

  • cohere.command-r-plus-v1:0: Most capable
  • cohere.command-r-v1:0: Balanced option

​
Model Selection Guide

Use CaseRecommended ModelWhy
Complex reasoningClaude Opus 4Best reasoning capabilities
Balanced performanceClaude Sonnet 3.5Speed + quality
Cost-effectiveNova LiteAmazon’s affordable option
Code generationClaude Sonnet 3.5Strong code understanding
High volumeLlama 3.1 8BFast and efficient

​
Best Practices

  1. Use Cross-Region Models: Better availability with us. prefix
  2. Enable Guardrails: For production content filtering
  3. Set Request Metadata: For tracking and debugging
  4. Choose Right Model: Match model to use case and budget
  5. Monitor CloudWatch: Track usage and costs
  6. Use IAM Roles: More secure than access keys
  7. Enable CloudTrail: Audit API calls
  8. Leverage Prompt Caching: For Claude models with repeated contexts

​
Security and Compliance

​
IAM Permissions

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

​
VPC Endpoints

Deploy Bedrock in VPC for enhanced security: 
import boto3

# Use VPC endpoint
client = boto3.client(
    'bedrock-runtime',
    endpoint_url='https://vpce-xxx.bedrock-runtime.region.vpce.amazonaws.com'
)

​
Compliance

Bedrock supports:
  • HIPAA
  • SOC 1, 2, 3
  • ISO 27001, 27017, 27018
  • PCI DSS
  • FedRAMP (select regions)

​
Cost Optimization

  1. Choose Appropriate Model: Don’t overpay for capabilities you don’t need
  2. Use Nova Models: Amazon’s models are cost-effective
  3. Enable Prompt Caching: Significant savings for Claude
  4. Monitor Usage: Set CloudWatch alarms for costs
  5. Use Lite/Micro Variants: For simple tasks