31.1.1 Bedrock 概述#

Amazon Bedrock 是 AWS 提供的完全托管服务，可以通过 API 访问基础模型，包括 Anthropic 的 Claude 模型。通过 Bedrock 使用 Claude Code 可以为企业带来以下优势：

Bedrock 的优势#

AWS 原生集成：与 AWS IAM、CloudTrail、CloudWatch 等服务无缝集成
企业级安全：符合 AWS 安全标准和合规要求
灵活的部署：支持多个 AWS 区域，满足数据驻留要求
成本管理：通过 AWS Cost Explorer 和 Budgets 进行成本控制
高可用性：利用 AWS 的全球基础设施和冗余机制

适用场景#

已经使用 AWS 基础设施的企业
需要符合特定数据驻留法规的组织
要求使用 AWS IAM 进行身份验证的场景
需要集中监控和日志记录的环境

31.1.2 Bedrock 配置步骤#

1. 前置条件检查#

class BedrockPrerequisitesChecker: """Bedrock 前置条件检查器"""

def init(self): self.checks = { 'aws_account': False, 'bedrock_enabled': False, 'model_access': False, 'iam_permissions': False, 'cli_configured': False }

def check_all(self) -> PrerequisiteReport: """检查所有前置条件""" report = PrerequisiteReport()

检查 AWS 账户

self.checks['aws_account'] = self._check_aws_account()

检查 Bedrock 是否启用

self.checks['bedrock_enabled'] = self._check_bedrock_enabled()

检查模型访问权限

self.checks['model_access'] = self._check_model_access()

检查 IAM 权限

self.checks['iam_permissions'] = self._check_iam_permissions()

检查 CLI 配置

self.checks['cli_configured'] = self._check_cli_configured()

生成报告

report.checks = self.checks report.all_passed = all(self.checks.values()) report.missing = [ check for check, passed in self.checks.items() if not passed ]

return report

def _check_aws_account(self) -> bool: """检查 AWS 账户""" try: result = subprocess.run( ['aws', 'sts', 'get-caller-identity'], capture_output=True,

text=True ) return result.returncode == 0 except Exception: return False

def _check_bedrock_enabled(self) -> bool: """检查 Bedrock 是否启用""" try: result = subprocess.run( ['aws', 'bedrock', 'list-foundation-models'], capture_output=True, text=True ) return result.returncode == 0 except Exception: return False

def _check_model_access(self) -> bool: """检查模型访问权限""" try: result = subprocess.run( ['aws', 'bedrock', 'list-inference-profiles'], capture_output=True, text=True ) return result.returncode == 0 except Exception: return False

2. 提交用例详情#

首次使用 Anthropic 模型需要提交用例详情：


bash
bash

# 通过 AWS CLI 提交
aws bedrock create-model-customization-job \
  --job-name "claude-code-use-case" \
  --base-model-identifier "anthropic.claude-sonnet-4-5-20250929-v1:0" \
  --customization-type "FINE_TUNING" \
  --customization-config '{
    "useCase": "Code generation and assistance",
    "teamSize": "10-50",
    "industry": "Technology"
  }'

### 3. 配置 AWS 凭证

#### 选项 A：AWS CLI 配置

# 配置 AWS CLI
aws configure
# 输入您的 AWS 凭证
AWS Access Key ID: [您的访问密钥]
AWS Secret Access Key: [您的秘密密钥]
Default region name: us-east-1
Default output format: json

选项 B：环境变量


bash
bash

# 设置环境变量
export AWS_ACCESS_KEY_ID=your-access-key-id
export AWS_SECRET_ACCESS_KEY=your-secret-access-key
export AWS_SESSION_TOKEN=your-session-token  # 如果使用临时凭证
export AWS_REGION=us-east-1

#### 选项 C：AWS SSO

# 配置 SSO 配置文件
aws configure sso
# 登录
aws sso login --profile claude-code
# 使用配置文件
export AWS_PROFILE=claude-code

选项 D：Bedrock API 密钥


bash
bash

# 使用 Bedrock API 密钥（推荐用于简化部署）
export AWS_BEARER_TOKEN_BEDROCK=your-bedrock-api-key

### 4. 启用 Claude Code Bedrock 集成

# 启用 Bedrock
export CLAUDE_CODE_USE_BEDROCK=1
# 设置区域
export AWS_REGION=us-east-1
# 可选：为小型/快速模型设置不同区域
export ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION=us-west-2

5. 配置模型#


bash
bash

# 主模型
export ANTHROPIC_MODEL='global.anthropic.claude-sonnet-4-5-20250929-v1:0'

# 小型/快速模型
export ANTHROPIC_SMALL_FAST_MODEL='us.anthropic.claude-haiku-4-5-20251001-v1:0'

# 使用推理配置文件
export ANTHROPIC_MODEL='arn:aws:bedrock:us-east-2:your-account-id:application-inference-profile/your-model-id'

## 31.1.3 IAM 权限配置

### 基础 IAM 策略

{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AllowModelAccess",
"Effect": "Allow",
"Action": [
"bedrock:InvokeModel",
"bedrock:InvokeModelWithResponseStream",
"bedrock:ListInferenceProfiles"
],
"Resource": [
"arn:aws:bedrock:*:*:inference-profile/*",
"arn:aws:bedrock:*:*:application-inference-profile/*",
"arn:aws:bedrock:*:*:foundation-model/*"
]
},
{
"Sid": "AllowMarketplaceAccess",
"Effect": "Allow",
"Action": [
"aws-marketplace:ViewSubscriptions",
"aws-marketplace:Subscribe"
],
"Resource": "*",
"Condition": {
"StringEquals": {
"aws:CalledViaLast": "bedrock.amazonaws.com"
}
}
}
]
}

严格 IAM 策略#


bash
json

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "AllowSpecificModelAccess",
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream"
      ],
      "Resource": [
        "arn:aws:bedrock:us-east-1:123456789012:inference-profile/global.anthropic.claude-sonnet-4-5-20250929-v1:0",
        "arn:aws:bedrock:us-west-2:123456789012:inference-profile/us.anthropic.claude-haiku-4-5-20251001-v1:0"
      ]
    }
  ]
}

### IAM 角色创建

# 创建 IAM 角色
aws iam create-role \
--role-name ClaudeCodeBedrockRole \
--assume-role-policy-document '{
"Version": "2012-10-17",
"Statement": [{
"Effect": "Allow",
"Principal": {"Service": "bedrock.amazonaws.com"},
"Action": "sts:AssumeRole"
}]
}'
# 附加策略
aws iam put-role-policy \
--role-name ClaudeCodeBedrockRole \
--policy-name ClaudeCodeBedrockPolicy \
--policy-document file://bedrock-policy.json

31.1.4 高级配置#

自动凭证刷新#


bash
json

{
  "awsAuthRefresh": "aws sso login --profile myprofile",
  "env": {
    "AWS_PROFILE": "myprofile",
    "CLAUDE_CODE_USE_BEDROCK": "1",
    "AWS_REGION": "us-east-1"
  }
}

### 输出令牌优化

# Bedrock 推荐的令牌设置
export CLAUDE_CODE_MAX_OUTPUT_TOKENS=4096
export MAX_THINKING_TOKENS=1024

提示缓存配置#


bash
bash

# 启用提示缓存（默认启用）
# 如需禁用
export DISABLE_PROMPT_CACHING=1

## 31.1.5 监控和故障排除

### CloudWatch 监控

class BedrockMonitor:
"""Bedrock 监控器"""
def __init__(self):
self.cloudwatch = boto3.client('cloudwatch')
self.metrics = [
'InvokeModel',
'InvokeModelWithResponseStream',
'Latency',
'ErrorCount',
'5XXError',
'4XXError'
]
def setup_alarms(self, config: Dict):
"""设置告警"""
for metric in self.metrics:
alarm = self._create_alarm(metric, config)
self.cloudwatch.put_metric_alarm(**alarm)
def _create_alarm(self, metric: str, config: Dict) -> Dict:
"""创建告警"""
return {
'AlarmName': f'Bedrock{metric}Alarm',
'MetricName': metric,
'Namespace': 'AWS/Bedrock',
'Statistic': 'Sum',
'Period': 300,
'EvaluationPeriods': 1,
'Threshold': config.get('threshold', 100),
'ComparisonOperator': 'GreaterThanThreshold',
'AlarmActions': [config.get('sns_topic')],
'TreatMissingData': 'notBreaching'
}

常见问题解决#


bash
python

class BedrockTroubleshooter:
    """Bedrock 故障排除器"""

    def diagnose(self, error: str) -> DiagnosisResult:
        """诊断问题"""
        if 'AccessDenied' in error:
            return self._diagnose_access_denied()
        elif 'ResourceNotFound' in error:
            return self._diagnose_resource_not_found()
        elif 'ThrottlingException' in error:
            return self._diagnose_throttling()
        elif 'ValidationException' in error:
            return self._diagnose_validation_error()
        else:
            return DiagnosisResult(
                issue='Unknown',
                solution='Check AWS CloudWatch logs for details'
            )

    def _diagnose_access_denied(self) -> DiagnosisResult:
        """诊断访问拒绝错误"""
        return DiagnosisResult(
            issue='IAM Permission Denied',
            solution='''1. Verify IAM user/role has bedrock:InvokeModel permission
2. Check if the model is accessible in the region
3. Verify the resource ARN in the IAM policy''',
            commands=[
                'aws iam get-role-policy --role-name <role-name> --policy-name <policy-name>',
                'aws bedrock list-inference-profiles --region us-east-1'
            ]
        )

    def _diagnose_throttling(self) -> DiagnosisResult:
        """诊断限流错误"""
        return DiagnosisResult(
            issue='Rate Limit Exceeded',
            solution='''1. Implement request queuing
2. Use inference profiles for higher throughput
3. Contact AWS support to increase quota''',
            commands=[
                'aws service-quotas list-service-quotas --service-code bedrock'
            ]
        )

通过正确配置 Amazon Bedrock，企业可以利用 AWS 的强大基础设施，安全、高效地部署 Claude Code。

Claude Code 深度教程

31.1 Amazon Bedrock 集成