推理提供者目标

借助推理提供者目标，您可以显式控制模型提供者的端点、模型映射和操作。当您需要自定义哪些模型可用、设置每个模型的令牌限制、配置路径重写或连接到没有内置连接器的提供程序时，请使用提供程序配置。

目标配置

推理提供者目标的目标配置使用以下结构：

{
    "inference": {
        "provider": {
            "endpoint": "https://api.openai.com",
            "operations": [
                {
                    "path": "/v1/chat/completions",
                    "models": [
                        {"model": "gpt-5.5"},
                        {"model": "gpt-5.4"},
                        {"model": "gpt-5.4-mini"}
                    ]
                },
                {
                    "path": "/v1/responses",
                    "models": [
                        {"model": "gpt-5.5"},
                        {"model": "gpt-5.4"}
                    ]
                }
            ]
        }
    }
}

创建提供者推理目标

以下示例使用提供程序配置创建 OpenAI 推理目标：

aws bedrock-agentcore-control create-gateway-target --cli-input-json '{
    "gatewayIdentifier": "GATEWAY_ID",
    "name": "openai",
    "targetConfiguration": {
        "inference": {
            "provider": {
                "endpoint": "https://api.openai.com",
                "operations": [
                    {
                        "path": "/v1/chat/completions",
                        "models": [
                            {"model": "gpt-5.5"},
                            {"model": "gpt-5.4"},
                            {"model": "gpt-5.4-mini"}
                        ]
                    },
                    {
                        "path": "/v1/responses",
                        "models": [
                            {"model": "gpt-5.5"},
                            {"model": "gpt-5.4"}
                        ]
                    }
                ]
            }
        }
    },
    "credentialProviderConfigurations": [
        {
            "credentialProviderType": "API_KEY",
            "credentialProvider": {
                "apiKeyCredentialProvider": {
                    "providerArn": "arn:aws:bedrock-agentcore:us-west-2:111122223333:token-vault/default/apikeycredentialprovider/openai-key",
                    "credentialLocation": "HEADER",
                    "credentialParameterName": "Authorization",
                    "credentialPrefix": "Bearer "
                }
            }
        }
    ]
}'

以下示例使用明确的提供者配置和模型映射创建一个 Bedrock 推理目标。带的modelMapping配置providerPrefix允许客户端使用简短的模型名称（比如claude-opus-4-7），而网关则将其转换为以提供者为前缀的名称（例如）：anthropic.claude-opus-4-7

aws bedrock-agentcore-control create-gateway-target --cli-input-json '{
    "gatewayIdentifier": "GATEWAY_ID",
    "name": "bedrock",
    "targetConfiguration": {
        "inference": {
            "provider": {
                "endpoint": "https://bedrock-mantle.us-east-1.api.aws",
                "modelMapping": {
                    "providerPrefix": {"strip": true, "separator": "."}
                },
                "operations": [
                    {
                        "path": "/v1/chat/completions",
                        "models": [
                            {"model": "anthropic.claude-opus-*"},
                            {"model": "anthropic.claude-sonnet-*"},
                            {"model": "openai.gpt-oss-*"}
                        ]
                    },
                    {
                        "path": "/v1/messages",
                        "providerPath": "/anthropic/v1/messages",
                        "models": [
                            {"model": "anthropic.claude-opus-*"},
                            {"model": "anthropic.claude-sonnet-*"}
                        ]
                    }
                ]
            }
        }
    },
    "credentialProviderConfigurations": [
        {"credentialProviderType": "GATEWAY_IAM_ROLE"}
    ]
}'

调用提供者推理目标

要调用推理目标，请向网关的/inference路径发送请求。网关根据请求正文中的model字段将每个请求路由到正确的目标。该model值可以是普通模型 ID（例如gpt-5.5），也可以是表单中的目标限定模型 ID{targetName}/{modelId}（例如）。openai/gpt-5.5有关如何将该model值与目标匹配的详细信息，请参阅Model-based 路由。

URL 格式如下：

https://{gatewayId}.gateway.bedrock-agentcore.{region}.amazonaws.com/inference/{path}

{path}替换为推理操作路径（例如v1/chat/completionsv1/responses、或v1/messages）。

使用 OpenAI 开发工具包

将网关的/inference/v1路径设置为base_url：

from openai import OpenAI

client = OpenAI(
    base_url="https://<gateway-id>.gateway.bedrock-agentcore.us-west-2.amazonaws.com/inference/v1",
    api_key="<gateway-auth-token>"
)

response = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "user", "content": "Hello!"}]
)

使用 Anthropic SDK

将网关的/inference路径设置为base_url：

import anthropic

client = anthropic.Anthropic(
    base_url="https://<gateway-id>.gateway.bedrock-agentcore.us-west-2.amazonaws.com/inference",
    api_key="<gateway-auth-token>"
)

response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}]
)

使用 awscurl

awscurl --service bedrock-agentcore --region us-west-2 -X POST \
    "https://<gateway-id>.gateway.bedrock-agentcore.us-west-2.amazonaws.com/inference/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -d '{"model": "gpt-5.5", "messages": [{"role": "user", "content": "Hello!"}]}'

合格的模型路由

当多个目标为同一个模型提供服务时，请在模型 ID 前面加上目标名称以路由到特定的提供商：

# Route explicitly to the "bedrock" target
response = client.chat.completions.create(
    model="bedrock/claude-opus-4-7",
    messages=[{"role": "user", "content": "Hello!"}]
)

Model-based 路由

网关根据请求正文中的model字段路由推理请求：

流式传输

直播遵循OpenAI SSE惯例。"stream": true在请求正文中设置，网关无需转换即可通过来自提供商的 SSE 流：

stream = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "user", "content": "Write a story."}],
    stream=True
)

for chunk in stream:
    print(chunk.choices[0].delta.content, end="")

出站授权

推理提供商目标支持以下出站授权类型：