本文属于机器翻译版本。若本译文内容与英语原文存在差异，则一律以英文原文为准。

压缩

Compaction 在接近上下文窗口限制时自动总结旧的上下文，从而延长了长时间运行的对话和任务的有效上下文长度。这非常适合：

以下型号支持压实：

压实的工作原理

启用压缩功能后，当对话接近配置的令牌阈值时，Claude会自动对其进行汇总。API：

在后续请求中，将回复附加到您的消息中。API 会自动删除屏蔽之前的所有消息compaction块，继续摘要中的对话。

基本用法

通过context_management.edits在您的消息 API 请求中添加compact_20260112策略来启用压缩。

CLI

aws bedrock-runtime invoke-model \
    --model-id "us.anthropic.claude-opus-4-6-v1" \
    --body '{
        "anthropic_version": "bedrock-2023-05-31",
        "anthropic_beta": ["compact-2026-01-12"],
        "max_tokens": 4096,
        "messages": [
            {
                "role": "user",
                "content": "Help me build a website"
            }
        ],
        "context_management": {
            "edits": [
                {
                    "type": "compact_20260112"
                }
            ]
        }
    }' \
    --cli-binary-format raw-in-base64-out \
    /tmp/response.json

echo "Response:"
cat /tmp/response.json | jq '.content[] | {type, text: .text[0:500]}'

Python

import boto3
import json

bedrock_runtime = boto3.client(service_name='bedrock-runtime')

messages = [{"role": "user", "content": "Help me build a website"}]

response = bedrock_runtime.invoke_model(
    modelId="us.anthropic.claude-opus-4-6-v1",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "anthropic_beta": ["compact-2026-01-12"],
        "max_tokens": 4096,
        "messages": messages,
        "context_management": {
            "edits": [
                {
                    "type": "compact_20260112"
                }
            ]
        }
    })
)

response_body = json.loads(response["body"].read())

# Append the response (including any compaction block) to continue the conversation
messages.append({"role": "assistant", "content": response_body["content"]})

for block in response_body["content"]:
    if block.get("type") == "compaction":
        print(f"[COMPACTION]: {block['content'][:200]}...")
    elif block.get("type") == "text":
        print(f"[RESPONSE]: {block['text']}")

TypeScript

import { BedrockRuntimeClient, InvokeModelCommand } from "@aws-sdk/client-bedrock-runtime";

async function main() {
    const client = new BedrockRuntimeClient({});

    const messages: Array<{role: string, content: string | object[]}> = [
        { role: "user", content: "Help me build a website" }
    ];

    const command = new InvokeModelCommand({
        modelId: "us.anthropic.claude-opus-4-6-v1",
        body: JSON.stringify({
            anthropic_version: "bedrock-2023-05-31",
            anthropic_beta: ["compact-2026-01-12"],
            max_tokens: 4096,
            messages,
            context_management: {
                edits: [
                    {
                        type: "compact_20260112"
                    }
                ]
            }
        })
    });

    const response = await client.send(command);
    const responseBody = JSON.parse(new TextDecoder().decode(response.body));

    // Append response to continue conversation
    messages.push({ role: "assistant", content: responseBody.content });

    for (const block of responseBody.content) {
        if (block.type === "compaction") {
            console.log(`[COMPACTION]: ${block.content.substring(0, 200)}...`);
        } else if (block.type === "text") {
            console.log(`[RESPONSE]: ${block.text}`);
        }
    }
}

main().catch(console.error);

参数

触发器配置

使用以下trigger参数配置何时触发压缩：

import boto3
import json

bedrock_runtime = boto3.client(service_name='bedrock-runtime')

response = bedrock_runtime.invoke_model(
    modelId="us.anthropic.claude-opus-4-6-v1",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "anthropic_beta": ["compact-2026-01-12"],
        "max_tokens": 4096,
        "messages": [{"role": "user", "content": "Help me build a website"}],
        "context_management": {
            "edits": [
                {
                    "type": "compact_20260112",
                    "trigger": {
                        "type": "input_tokens",
                        "value": 100000
                    }
                }
            ]
        }
    })
)

response_body = json.loads(response["body"].read())
print(response_body["content"][-1]["text"])

自定义摘要说明

默认情况下，compaction 使用以下摘要提示：

You have written a partial transcript for the initial task above. Please write a summary of the transcript. The purpose of this summary is to provide continuity so you can continue to make progress towards solving the task in a future context, where the raw history above may not be accessible and will be replaced with this summary. Write down anything that would be helpful, including the state, next steps, learnings etc. You must wrap your summary in a <summary></summary> block.

您可以通过instructions参数提供自定义指令来完全替换此提示。自定义说明并不能补充默认指令；它们完全取代了默认指令：

import boto3
import json

bedrock_runtime = boto3.client(service_name='bedrock-runtime')

response = bedrock_runtime.invoke_model(
    modelId="us.anthropic.claude-opus-4-6-v1",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "anthropic_beta": ["compact-2026-01-12"],
        "max_tokens": 4096,
        "messages": [{"role": "user", "content": "Help me build a website"}],
        "context_management": {
            "edits": [
                {
                    "type": "compact_20260112",
                    "instructions": "Focus on preserving code snippets, variable names, and technical decisions."
                }
            ]
        }
    })
)

response_body = json.loads(response["body"].read())
print(response_body["content"][-1]["text"])

压实后暂停

用于pause_after_compaction在生成压缩摘要后暂停 API。这允许您在 API 继续响应之前添加其他内容块（例如保留最近的消息或特定的面向指令的消息）。

启用后，API 将在生成压缩块后返回一条包含compaction停止原因的消息：

import boto3
import json

bedrock_runtime = boto3.client(service_name='bedrock-runtime')

messages = [{"role": "user", "content": "Help me build a website"}]

response = bedrock_runtime.invoke_model(
    modelId="us.anthropic.claude-opus-4-6-v1",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "anthropic_beta": ["compact-2026-01-12"],
        "max_tokens": 4096,
        "messages": messages,
        "context_management": {
            "edits": [
                {
                    "type": "compact_20260112",
                    "pause_after_compaction": True
                }
            ]
        }
    })
)

response_body = json.loads(response["body"].read())

# Check if compaction triggered a pause
if response_body.get("stop_reason") == "compaction":
    # Response contains only the compaction block
    messages.append({"role": "assistant", "content": response_body["content"]})

    # Continue the request
    response = bedrock_runtime.invoke_model(
        modelId="us.anthropic.claude-opus-4-6-v1",
        body=json.dumps({
            "anthropic_version": "bedrock-2023-05-31",
            "anthropic_beta": ["compact-2026-01-12"],
            "max_tokens": 4096,
            "messages": messages,
            "context_management": {
                "edits": [{"type": "compact_20260112"}]
            }
        })
    )
    response_body = json.loads(response["body"].read())

print(response_body["content"][-1]["text"])

使用压实块

当触发压缩时，API 会在助手响应开始时返回一个compaction区块。

长时间进行的对话可能会导致多次压缩。最后一个压缩块反映了提示的最终状态，将之前的内容替换为生成的摘要。

{
  "content": [
    {
      "type": "compaction",
      "content": "Summary of the conversation: The user requested help building a web scraper..."
    },
    {
      "type": "text",
      "text": "Based on our conversation so far..."
    }
  ]
}

流式传输

启用压缩功能的流式传输响应时，您将在压缩开始时收到一个content_block_start事件。压缩块的流式传输方式与文本块不同。您将收到一个content_block_start活动，然后是包含完整摘要内容content_block_delta的单曲（没有中间直播），然后是一个content_block_stop活动。

提示缓存

你可以在 compaction blocks 上添加一个cache_control断点，它会缓存完整的系统提示以及摘要内容。原始压缩内容将被忽略。请注意，当触发压缩时，可能会导致后续请求出现缓存失误。

{
    "role": "assistant",
    "content": [
        {
            "type": "compaction",
            "content": "[summary text]",
            "cache_control": {"type": "ephemeral"}
        },
        {
            "type": "text",
            "text": "Based on our conversation..."
        }
    ]
}

了解使用情况

压缩需要额外的采样步骤，这会导致费率限制和计费。API 在响应中返回详细的使用信息：

{
  "usage": {
    "input_tokens": 45000,
    "output_tokens": 1234,
    "iterations": [
      {
        "type": "compaction",
        "input_tokens": 180000,
        "output_tokens": 3500
      },
      {
        "type": "message",
        "input_tokens": 23000,
        "output_tokens": 1000
      }
    ]
  }
}

该iterations数组显示每次采样迭代的使用情况。当压缩发生时，你会看到一个compaction迭代，然后是主message迭代。最后一次迭代的代币数量反映了压缩后的有效上下文大小。