流式处理

Deep Agents 建立在 LangGraph 的流式基础设施之上，对子代理流提供一流支持。当深度代理将工作委托给子代理时，您可以独立地从每个子代理流式传输更新——实时跟踪进度、LLM 令牌和工具调用。深度代理流式处理能实现什么：

流式传输子代理进度—跟踪每个子代理并行运行时的执行情况。
流式传输 LLM 令牌—从主代理和每个子代理流式传输令牌。
流式传输工具调用—查看子代理执行过程中的工具调用和结果。
流式传输自定义更新—从子代理节点内部发出用户定义的信号。

启用子图流式传输

Deep Agents 使用 LangGraph 的子图流式传输来呈现来自子代理执行的事件。要接收子代理事件，请在流式传输时启用 stream_subgraphs。

import { createDeepAgent } from "deepagents";

const agent = createDeepAgent({
  systemPrompt: "You are a helpful research assistant",
  subagents: [
    {
      name: "researcher",
      description: "Researches a topic in depth",
      systemPrompt: "You are a thorough researcher.",
    },
  ],
});

for await (const [namespace, chunk] of await agent.stream(
  { messages: [{ role: "user", content: "Research quantum computing advances" }] },
  {
    streamMode: "updates",
    subgraphs: true,
  }
)) {
  if (namespace.length > 0) {
    // 子代理事件 - 命名空间标识来源
    console.log(`[subagent: ${namespace.join("|")}]`);
  } else {
    // 主代理事件
    console.log("[main agent]");
  }
  console.log(chunk);
}

命名空间

当启用 subgraphs 时，每个流式事件都包含一个命名空间，用于标识是哪个代理产生的。命名空间是节点名称和任务 ID 的路径，代表代理层次结构。

命名空间	来源
`()` (空)	主代理
`("tools:abc123",)`	由主代理的 `task` 工具调用 `abc123` 生成的子代理
`("tools:abc123", "model_request:def456")`	子代理内部的模型请求节点

使用命名空间将事件路由到正确的 UI 组件：

for await (const [namespace, chunk] of await agent.stream(
  { messages: [{ role: "user", content: "Plan my vacation" }] },
  { streamMode: "updates", subgraphs: true }
)) {
  // 检查此事件是否来自子代理
  const isSubagent = namespace.some(
    (segment: string) => segment.startsWith("tools:")
  );

  if (isSubagent) {
    // 从命名空间中提取工具调用 ID
    const toolCallId = namespace
      .find((s: string) => s.startsWith("tools:"))
      ?.split(":")[1];
    console.log(`Subagent ${toolCallId}:`, chunk);
  } else {
    console.log("Main agent:", chunk);
  }
}

子代理进度

使用 stream_mode="updates" 在每个步骤完成时跟踪子代理进度。这对于显示哪些子代理处于活动状态以及它们完成了哪些工作非常有用。

import { createDeepAgent } from "deepagents";

const agent = createDeepAgent({
  systemPrompt:
    "You are a project coordinator. Always delegate research tasks " +
    "to your researcher subagent using the task tool. Keep your final response to one sentence.",
  subagents: [
    {
      name: "researcher",
      description: "Researches topics thoroughly",
      systemPrompt:
        "You are a thorough researcher. Research the given topic " +
        "and provide a concise summary in 2-3 sentences.",
    },
  ],
});

for await (const [namespace, chunk] of await agent.stream(
  {
    messages: [
      { role: "user", content: "Write a short summary about AI safety" },
    ],
  },
  { streamMode: "updates", subgraphs: true },
)) {
  // 主代理更新（空命名空间）
  if (namespace.length === 0) {
    for (const [nodeName, data] of Object.entries(chunk)) {
      if (nodeName === "tools") {
        // 返回给主代理的子代理结果
        for (const msg of (data as any).messages ?? []) {
          if (msg.type === "tool") {
            console.log(`\nSubagent complete: ${msg.name}`);
            console.log(`  Result: ${String(msg.content).slice(0, 200)}...`);
          }
        }
      } else {
        console.log(`[main agent] step: ${nodeName}`);
      }
    }
  }
  // 子代理更新（非空命名空间）
  else {
    for (const [nodeName] of Object.entries(chunk)) {
      console.log(`  [${namespace[0]}] step: ${nodeName}`);
    }
  }
}

输出

Main agent step: model_request
  [tools:call_abc123] step: model_request
  [tools:call_abc123] step: tools
  [tools:call_abc123] step: model_request
Subagent complete: task
Result: ## AI Safety Report...
Main agent step: model_request
  [tools:call_def456] step: model_request
  [tools:call_def456] step: model_request
Subagent complete: task
Result: # Comprehensive Report on AI Safety...
Main agent step: model_request

LLM 令牌

使用 stream_mode="messages" 从主代理和子代理流式传输单个令牌。每个消息事件都包含标识来源代理的元数据。

let currentSource = "";

for await (const [namespace, chunk] of await agent.stream(
  {
    messages: [
      {
        role: "user",
        content: "Research quantum computing advances",
      },
    ],
  },
  { streamMode: "messages", subgraphs: true },
)) {
  const [message] = chunk;

  // 检查此事件是否来自子代理（命名空间包含 "tools:"）
  const isSubagent = namespace.some((s: string) => s.startsWith("tools:"));

  if (isSubagent) {
    // 来自子代理的令牌
    const subagentNs = namespace.find((s: string) => s.startsWith("tools:"))!;
    if (subagentNs !== currentSource) {
      process.stdout.write(`\n\n--- [subagent: ${subagentNs}] ---\n`);
      currentSource = subagentNs;
    }
    if (message.text) {
      process.stdout.write(message.text);
    }
  } else {
    // 来自主代理的令牌
    if ("main" !== currentSource) {
      process.stdout.write(`\n\n--- [main agent] ---\n`);
      currentSource = "main";
    }
    if (message.text) {
      process.stdout.write(message.text);
    }
  }
}

process.stdout.write("\n");

工具调用

当子代理使用工具时，您可以流式传输工具调用事件以显示每个子代理正在做什么。工具调用块出现在 messages 流模式中。

import { AIMessageChunk, ToolMessage } from "langchain";

for await (const [namespace, chunk] of await agent.stream(
  {
    messages: [
      {
        role: "user",
        content: "Research recent quantum computing advances",
      },
    ],
  },
  { streamMode: "messages", subgraphs: true },
)) {
  const [message] = chunk;

  // 识别来源："main" 或子代理命名空间段
  const isSubagent = namespace.some((s: string) => s.startsWith("tools:"));
  const source = isSubagent
    ? namespace.find((s: string) => s.startsWith("tools:"))!
    : "main";

  // 工具调用块（流式工具调用）
  if (AIMessageChunk.isInstance(message) && message.tool_call_chunks?.length) {
    for (const tc of message.tool_call_chunks) {
      if (tc.name) {
        console.log(`\n[${source}] Tool call: ${tc.name}`);
      }
      // 参数以块形式流式传输 - 增量写入
      if (tc.args) {
        process.stdout.write(tc.args);
      }
    }
  }

  // 工具结果
  if (ToolMessage.isInstance(message)) {
    console.log(
      `\n[${source}] Tool result [${message.name}]: ${message.text?.slice(0, 150)}`,
    );
  }

  // 常规 AI 内容（跳过工具调用消息）
  if (
    AIMessageChunk.isInstance(message) &&
    message.text &&
    !message.tool_call_chunks?.length
  ) {
    process.stdout.write(message.text);
  }
}

process.stdout.write("\n");

自定义更新

在子代理工具内使用 config.writer 发出自定义进度事件：

import { createDeepAgent } from "deepagents";
import { tool, type ToolRuntime } from "langchain";
import { z } from "zod";

/**
 * 一个通过 config.writer 发出自定义进度事件的工具。
 * writer 将数据发送到 "custom" 流模式。
 */
const analyzeData = tool(
  async ({ topic }: { topic: string }, config: ToolRuntime) => {
    const writer = config.writer;

    writer?.({ status: "starting", topic, progress: 0 });
    await new Promise((r) => setTimeout(r, 500));

    writer?.({ status: "analyzing", progress: 50 });
    await new Promise((r) => setTimeout(r, 500));

    writer?.({ status: "complete", progress: 100 });
    return `Analysis of "${topic}": Customer sentiment is 85% positive, driven by product quality and support response times.`;
  },
  {
    name: "analyze_data",
    description:
      "Run a data analysis on a given topic. " +
      "This tool performs the actual analysis and emits progress updates. " +
      "You MUST call this tool for any analysis request.",
    schema: z.object({
      topic: z.string().describe("The topic or subject to analyze"),
    }),
  },
);

const agent = createDeepAgent({
  systemPrompt:
    "You are a coordinator. For any analysis request, you MUST delegate " +
    "to the analyst subagent using the task tool. Never try to answer directly. " +
    "After receiving the result, summarize it in one sentence.",
  subagents: [
    {
      name: "analyst",
      description: "Performs data analysis with real-time progress tracking",
      systemPrompt:
        "You are a data analyst. You MUST call the analyze_data tool " +
        "for every analysis request. Do not use any other tools. " +
        "After the analysis completes, report the result.",
      tools: [analyzeData],
    },
  ],
});

for await (const [namespace, chunk] of await agent.stream(
  {
    messages: [
      {
        role: "user",
        content: "Analyze customer satisfaction trends",
      },
    ],
  },
  { streamMode: "custom", subgraphs: true },
)) {
  const isSubagent = namespace.some((s: string) => s.startsWith("tools:"));
  if (isSubagent) {
    const subagentNs = namespace.find((s: string) => s.startsWith("tools:"))!;
    console.log(`[${subagentNs}]`, chunk);
  } else {
    console.log("[main]", chunk);
  }
}

输出

[tools:call_abc123] { status: 'fetching', progress: 0 }
[tools:call_abc123] { status: 'analyzing', progress: 50 }
[tools:call_abc123] { status: 'complete', progress: 100 }

流式传输多种模式

组合多种流模式以获取代理执行的完整视图：

// 跳过内部中间件步骤 - 仅显示有意义的节点名称
const INTERESTING_NODES = new Set(["model_request", "tools"]);

let lastSource = "";
let midLine = false; // 当我们写入了令牌但没有尾随换行符时为 true

for await (const [namespace, mode, data] of await agent.stream(
  {
    messages: [
      {
        role: "user",
        content: "Analyze the impact of remote work on team productivity",
      },
    ],
  },
  { streamMode: ["updates", "messages", "custom"], subgraphs: true },
)) {
  const isSubagent = namespace.some((s: string) => s.startsWith("tools:"));
  const source = isSubagent ? "subagent" : "main";

  if (mode === "updates") {
    for (const nodeName of Object.keys(data)) {
      if (!INTERESTING_NODES.has(nodeName)) continue;
      if (midLine) {
        process.stdout.write("\n");
        midLine = false;
      }
      console.log(`[${source}] step: ${nodeName}`);
    }
  } else if (mode === "messages") {
    const [message] = data;
    if (message.text) {
      // 当来源更改时打印标题
      if (source !== lastSource) {
        if (midLine) {
          process.stdout.write("\n");
          midLine = false;
        }
        process.stdout.write(`\n[${source}] `);
        lastSource = source;
      }
      process.stdout.write(message.text);
      midLine = true;
    }
  } else if (mode === "custom") {
    if (midLine) {
      process.stdout.write("\n");
      midLine = false;
    }
    console.log(`[${source}] custom event:`, data);
  }
}

process.stdout.write("\n");

常见模式

跟踪子代理生命周期

监控子代理何时启动、运行和完成：

for await (const [namespace, chunk] of await agent.stream(
  {
    messages: [
      { role: "user", content: "Research the latest AI safety developments" },
    ],
  },
  { streamMode: "updates", subgraphs: true },
)) {
  for (const [nodeName, data] of Object.entries(chunk)) {
    // ─── 阶段 1：检测子代理启动 ────────────────────────
    // 当主代理的 model_request 包含 task 工具调用时，
    // 子代理已被生成。
    if (namespace.length === 0 && nodeName === "model_request") {
      for (const msg of (data as any).messages ?? []) {
        for (const tc of msg.tool_calls ?? []) {
          if (tc.name === "task") {
            activeSubagents.set(tc.id, {
              type: tc.args?.subagent_type,
              description: tc.args?.description?.slice(0, 80),
              status: "pending",
            });
            console.log(
              `[lifecycle] PENDING  → subagent "${tc.args?.subagent_type}" (${tc.id})`,
            );
          }
        }
      }
    }

    // ─── 阶段 2：检测子代理运行 ─────────────────────────
    // 当我们收到来自 tools:UUID 命名空间的事件时，
    // 该子代理正在主动执行。
    if (namespace.length > 0 && namespace[0].startsWith("tools:")) {
      const pregelId = namespace[0].split(":")[1];
      // 检查是否有任何待处理的子代理需要标记为运行中。
      // 注意：pregel 任务 ID 与 tool_call_id 不同，
      // 因此我们在第一个子代理事件时将任何待处理的子代理标记为运行中。
      for (const [id, sub] of activeSubagents) {
        if (sub.status === "pending") {
          sub.status = "running";
          console.log(
            `[lifecycle] RUNNING  → subagent "${sub.type}" (pregel: ${pregelId})`,
          );
          break;
        }
      }
    }

    // ─── 阶段 3：检测子代理完成 ──────────────────────
    // 当主代理的 tools 节点返回工具消息时，
    // 子代理已完成并返回了其结果。
    if (namespace.length === 0 && nodeName === "tools") {
      for (const msg of (data as any).messages ?? []) {
        if (msg.type === "tool") {
          const subagent = activeSubagents.get(msg.tool_call_id);
          if (subagent) {
            subagent.status = "complete";
            console.log(
              `[lifecycle] COMPLETE → subagent "${subagent.type}" (${msg.tool_call_id})`,
            );
            console.log(
              `  Result preview: ${String(msg.content).slice(0, 120)}...`,
            );
          }
        }
      }
    }
  }
}

// 打印最终状态
console.log("\n--- Final subagent states ---");
for (const [id, sub] of activeSubagents) {
  console.log(`  ${sub.type}: ${sub.status}`);
}

Get started

Deployment

Core capabilities

Frontend

Protocols

Command line interface

启用子图流式传输

命名空间

子代理进度

LLM 令牌

工具调用

自定义更新

流式传输多种模式

常见模式

跟踪子代理生命周期

相关内容

​启用子图流式传输

​命名空间

​子代理进度

​LLM 令牌

​工具调用

​自定义更新

​流式传输多种模式

​常见模式

​跟踪子代理生命周期

​相关内容

启用子图流式传输

命名空间

子代理进度

LLM 令牌

工具调用

自定义更新

流式传输多种模式

常见模式

跟踪子代理生命周期

相关内容