ChatOpenAI 集成

使用 LangChain JavaScript 与 ChatOpenAI 聊天模型集成。

OpenAI 是一家人工智能（AI）研究实验室。

本指南将帮助您开始使用 OpenAI 聊天模型。有关所有 ChatOpenAI 功能和配置的详细文档，请前往 API 参考。

Chat 完成 API 兼容性

ChatOpenAI 完全兼容 OpenAI 的（遗留）Chat 完成 API。如果您正在寻找连接到支持 Chat 完成 API 的其他模型提供商的方法，您可以这样做——参见 说明。

托管在 Azure 上的 OpenAI 模型

请注意，某些 OpenAI 模型也可以通过 Microsoft Azure 平台 访问。

概述

集成详情

类	包	可序列化	Python 支持	下载量	版本
ChatOpenAI	@langchain/openai	✅	✅

模型功能

请参阅下表表头中的链接，以获取如何使用特定功能的指南。

工具调用	结构化输出	图像输入	音频输入	视频输入	Token 级流式传输	Token 用量	Logprobs
✅	✅	✅	❌	❌	✅	✅	✅

设置

要访问 OpenAI 聊天模型，您需要创建一个 OpenAI 账户，获取 API 密钥，并安装 @langchain/openai 集成包。

凭据

前往 OpenAI 网站 注册 OpenAI 并生成 API 密钥。完成后，设置 OPENAI_API_KEY 环境变量：

export OPENAI_API_KEY="your-api-key"

如果您希望自动追踪您的模型调用，您还可以通过取消注释以下内容来设置您的 LangSmith API 密钥：

# export LANGSMITH_TRACING="true"
# export LANGSMITH_API_KEY="your-api-key"

安装

LangChain [ChatOpenAI] 集成位于 @langchain/openai 包中：

npm

npm install @langchain/openai @langchain/core

实例化

现在我们可以实例化我们的模型对象并生成聊天补全：

import { ChatOpenAI } from "@langchain/openai"

const llm = new ChatOpenAI({
  model: "gpt-4.1",
  temperature: 0,
  // other params...
})

调用

const aiMsg = await llm.invoke([
  {
    role: "system",
    content: "You are a helpful assistant that translates English to French. Translate the user sentence.",
  },
  {
    role: "user",
    content: "I love programming."
  },
])
aiMsg

AIMessage {
  "id": "chatcmpl-ADItECqSPuuEuBHHPjeCkh9wIO1H5",
  "content": "J'adore la programmation.",
  "additional_kwargs": {},
  "response_metadata": {
    "tokenUsage": {
      "completionTokens": 5,
      "promptTokens": 31,
      "totalTokens": 36
    },
    "finish_reason": "stop",
    "system_fingerprint": "fp_5796ac6771"
  },
  "tool_calls": [],
  "invalid_tool_calls": [],
  "usage_metadata": {
    "input_tokens": 31,
    "output_tokens": 5,
    "total_tokens": 36
  }
}

console.log(aiMsg.content)

J'adore la programmation.

自定义 URL

您可以通过传递 configuration 参数来自定义 SDK 发送请求的基础 URL，如下所示：

import { ChatOpenAI } from "@langchain/openai";

const llmWithCustomURL = new ChatOpenAI({
  model: "gpt-4.1",
  temperature: 0.9,
  configuration: {
    baseURL: "https://your_custom_url.com",
  },
});

await llmWithCustomURL.invoke("Hi there!");

configuration 字段还接受官方 SDK 接受的其它 ClientOptions 参数。

如果您在 Azure OpenAI 上托管，请参阅 专用页面。

自定义标头

您可以在相同的 configuration 字段中指定自定义标头：

import { ChatOpenAI } from "@langchain/openai";

const llmWithCustomHeaders = new ChatOpenAI({
  model: "gpt-4.1",
  temperature: 0.9,
  configuration: {
    defaultHeaders: {
      "Authorization": `Bearer SOME_CUSTOM_VALUE`,
    },
  },
});

await llmWithCustomHeaders.invoke("Hi there!");

禁用流式传输使用情况元数据

一些代理或第三方提供商提供与 OpenAI 大致相同的 API 接口，但不支持较新添加的 stream_options 参数来返回流式使用情况。您可以使用 ChatOpenAI 通过以下方式禁用流式传输来访问这些提供商：

import { ChatOpenAI } from "@langchain/openai";

const llmWithoutStreamUsage = new ChatOpenAI({
  model: "gpt-4.1",
  temperature: 0.9,
  streamUsage: false,
  configuration: {
    baseURL: "https://proxy.com",
  },
});

await llmWithoutStreamUsage.invoke("Hi there!");

调用微调模型

您可以通过传递相应的 modelName 参数来调用微调后的 OpenAI 模型。

这通常采用 ft:{OPENAI_MODEL_NAME}:{ORG_NAME}::{MODEL_ID} 的形式。例如：

import { ChatOpenAI } from "@langchain/openai";

const fineTunedLlm = new ChatOpenAI({
  temperature: 0.9,
  model: "ft:gpt-3.5-turbo-0613:{ORG_NAME}::{MODEL_ID}",
});

await fineTunedLlm.invoke("Hi there!");

生成元数据

如果您需要额外信息，如 logprobs 或 token 用量，这些信息将直接作为 response_metadata 字段中的内容返回在 invoke 响应的消息中。

需要 @langchain/core 版本 >=0.1.48。

import { ChatOpenAI } from "@langchain/openai";

// See https://cookbook.openai.com/examples/using_logprobs for details
const llmWithLogprobs = new ChatOpenAI({
  model: "gpt-4.1",
  logprobs: true,
  // topLogprobs: 5,
});

const responseMessageWithLogprobs = await llmWithLogprobs.invoke("Hi there!");
console.dir(responseMessageWithLogprobs.response_metadata.logprobs, { depth: null });

{
  content: [
    {
      token: 'Hello',
      logprob: -0.0004740447,
      bytes: [ 72, 101, 108, 108, 111 ],
      top_logprobs: []
    },
    {
      token: '!',
      logprob: -0.00004334534,
      bytes: [ 33 ],
      top_logprobs: []
    },
    {
      token: ' How',
      logprob: -0.000030113732,
      bytes: [ 32, 72, 111, 119 ],
      top_logprobs: []
    },
    {
      token: ' can',
      logprob: -0.0004797665,
      bytes: [ 32, 99, 97, 110 ],
      top_logprobs: []
    },
    {
      token: ' I',
      logprob: -7.89631e-7,
      bytes: [ 32, 73 ],
      top_logprobs: []
    },
    {
      token: ' assist',
      logprob: -0.114006,
      bytes: [
         32,  97, 115,
        115, 105, 115,
        116
      ],
      top_logprobs: []
    },
    {
      token: ' you',
      logprob: -4.3202e-7,
      bytes: [ 32, 121, 111, 117 ],
      top_logprobs: []
    },
    {
      token: ' today',
      logprob: -0.00004501419,
      bytes: [ 32, 116, 111, 100, 97, 121 ],
      top_logprobs: []
    },
    {
      token: '?',
      logprob: -0.000010206721,
      bytes: [ 63 ],
      top_logprobs: []
    }
  ],
  refusal: null
}

自定义工具

自定义工具 支持具有任意字符串输入的工具。当您预期字符串参数较长或复杂时，它们特别有用。

如果您使用的模型支持自定义工具，可以使用 ChatOpenAI 类和 customTool 函数来创建自定义工具。

import { ChatOpenAI, customTool } from "@langchain/openai";
import { createAgent, HumanMessage } from "langchain";

const codeTool = customTool(
  async () => {
    // ... Add code to execute the input
    return "Code executed successfully";
  },
  {
    name: "execute_code",
    description: "Execute a code snippet",
    format: { type: "text" },
  }
);

const model = new ChatOpenAI({ model: "gpt-5" });

const agent = createAgent({
  model,
  tools: [codeTool],
});

const result = await agent.invoke({
  messages: [new HumanMessage("Use the tool to execute the code")],
});
console.log(result);

strict: true

自 2024 年 8 月 6 日起，OpenAI 支持在调用工具时使用 strict 参数，这将强制模型遵守工具参数架构。查看更多。

需要 @langchain/openai >= 0.2.6

如果 strict: true，工具定义也将被验证，并且只接受 JSON schema 的子集。关键的是，架构不能有可选参数（那些带有默认值的参数）。阅读 完整文档 了解支持的架构类型。

这是一个带有工具调用的示例。向 .bindTools 传递额外的 strict: true 参数会将该参数传递给所有工具定义：

import { ChatOpenAI } from "@langchain/openai";
import { tool } from "@langchain/core/tools";
import * as z from "zod";

const weatherTool = tool((_) => "no-op", {
  name: "get_current_weather",
  description: "Get the current weather",
  schema: z.object({
    location: z.string(),
  }),
})

const llmWithStrictTrue = new ChatOpenAI({
  model: "gpt-4.1",
}).bindTools([weatherTool], {
  strict: true,
  tool_choice: weatherTool.name,
});

// Although the question is not about the weather, it will call the tool with the correct arguments
// because we passed `tool_choice` and `strict: true`.
const strictTrueResult = await llmWithStrictTrue.invoke("What is 127862 times 12898 divided by 2?");

console.dir(strictTrueResult.tool_calls, { depth: null });

[
  {
    name: 'get_current_weather',
    args: { location: 'current' },
    type: 'tool_call',
    id: 'call_hVFyYNRwc6CoTgr9AQFQVjm9'
  }
]

如果您只想将此参数应用于选定的工具，您也可以直接传递 OpenAI 格式的工具架构：

import { zodToJsonSchema } from "zod-to-json-schema";

const toolSchema = {
  type: "function",
  function: {
    name: "get_current_weather",
    description: "Get the current weather",
    strict: true,
    parameters: zodToJsonSchema(
      z.object({
        location: z.string(),
      })
    ),
  },
};

const llmWithStrictTrueTools = new ChatOpenAI({
  model: "gpt-4.1",
}).bindTools([toolSchema], {
  strict: true,
});

const weatherToolResult = await llmWithStrictTrueTools.invoke([{
  role: "user",
  content: "What is the current weather in London?"
}])

weatherToolResult.tool_calls;

[
  {
    name: 'get_current_weather',
    args: { location: 'London' },
    type: 'tool_call',
    id: 'call_EOSejtax8aYtqpchY8n8O82l'
  }
]

结构化输出

我们也可以将 strict: true 传递给 .withStructuredOutput()。这是一个示例：

import { ChatOpenAI } from "@langchain/openai";

const traitSchema = z.object({
  traits: z.array(z.string()).describe("A list of traits contained in the input"),
});

const structuredLlm = new ChatOpenAI({
  model: "gpt-4.1-mini",
}).withStructuredOutput(traitSchema, {
  name: "extract_traits",
  strict: true,
});

await structuredLlm.invoke([{
  role: "user",
  content: `I am 6'5" tall and love fruit.`
}]);

{ traits: [ `6'5" tall`, 'love fruit' ] }

响应 API

兼容性

以下要点适用于 @langchain/openai>=0.4.5-rc.0。

OpenAI 支持一个 响应 API，旨在构建 智能体 应用程序。它包括一套 内置工具，包括网页和文件搜索。它还支持 对话状态 的管理，允许您继续对话线程而无需显式传递之前的消息。

如果使用其中任一功能，ChatOpenAI 将路由到响应 API。也可以在实例化 ChatOpenAI 时指定 useResponsesApi: true。

内置工具

为 ChatOpenAI 配备内置工具将使其响应基于外部信息，例如通过文件或网页中的上下文。模型生成的 AIMessage 将包含有关内置工具调用的信息。

网页搜索

要触发网页搜索，像对待其他工具一样，向模型传递 {"type": "web_search_preview"}。

您也可以将内置工具作为调用参数传递：

llm.invoke("...", { tools: [{ type: "web_search_preview" }] });

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({ model: "gpt-4.1-mini" }).bindTools([
  { type: "web_search_preview" },
]);

await llm.invoke("What was a positive news story from today?");

注意，响应包含结构化的 内容块，其中包括响应文本和 OpenAI 引用 来源。输出消息还将包含任何工具调用的信息。

文件搜索

要触发文件搜索，像对待其他工具一样，向模型传递 文件搜索工具。您需要填充 OpenAI 管理的向量存储，并在工具定义中包含向量存储 ID。详见 OpenAI 文档。

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({ model: "gpt-4.1-mini" }).bindTools([
  { type: "file_search", vector_store_ids: ["vs..."] },
]);

await llm.invoke("Is deep research by OpenAI?");

与 网页搜索 一样，响应将包含带有引用的内容块。它还将包含来自内置工具调用的信息。

计算机使用

ChatOpenAI 支持 computer-use-preview 模型，这是用于内置计算机使用工具的专用模型。要启用，像对待其他工具一样传递 计算机使用工具。

目前，计算机使用的工具输出存在于 AIMessage.additional_kwargs.tool_outputs 中。要回复计算机使用工具调用，需要在创建相应的 ToolMessage 时设置 additional_kwargs.type: "computer_call_output"。

详见 OpenAI 文档。

import { AIMessage, ToolMessage } from "@langchain/core/messages";
import { ChatOpenAI } from "@langchain/openai";
import * as fs from "node:fs/promises";

const findComputerCall = (message: AIMessage) => {
  const toolOutputs = message.additional_kwargs.tool_outputs as
    | { type: "computer_call"; call_id: string; action: { type: string } }[]
    | undefined;

  return toolOutputs?.find((toolOutput) => toolOutput.type === "computer_call");
};

const llm = new ChatOpenAI({ model: "computer-use-preview" })
  .bindTools([
    {
      type: "computer-preview",
      display_width: 1024,
      display_height: 768,
      environment: "browser",
    },
  ])
  .bind({ truncation: "auto" });

let message = await llm.invoke("Check the latest OpenAI news on bing.com.");
const computerCall = findComputerCall(message);

if (computerCall) {
  // Act on a computer call action
  const screenshot = await fs.readFile("./screenshot.png", {
    encoding: "base64",
  });

  message = await llm.invoke(
    [
      new ToolMessage({
        additional_kwargs: { type: "computer_call_output" },
        tool_call_id: computerCall.call_id,
        content: [
          {
            type: "computer_screenshot",
            image_url: `data:image/png;base64,${screenshot}`,
          },
        ],
      }),
    ],
    { previous_response_id: message.response_metadata["id"] }
  );
}

代码解释器

ChatOpenAI 允许您使用内置的 代码解释器工具 来支持沙盒代码生成和执行。

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({
  model: "o4-mini",
  useResponsesApi: true,
});

const llmWithTools = llm.bindToools([
  {
    type: "code_interpreter",
    // Creates a new container
    container: { type: "auto" }
  },
]);

const response = await llmWithTools.invoke(
  "Write and run code to answer the question: what is 3^3?"
);

注意，上述命令创建了一个新的 容器。我们可以通过指定现有容器 ID 在调用之间重用容器。

const tool_outputs: Record<string, any>[] = response.additional_kwargs.tool_outputs
const container_id = tool_outputs[0].container_id

const llmWithTools = llm.bindTools([
  {
    type: "code_interpreter",
    // Reuses container from the last call
    container: container_id,
  },
]);

远程 MCP

ChatOpenAI 支持内置的 远程 MCP 工具，允许在 OpenAI 服务器上发生由模型生成的对 MCP 服务器的调用。

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({
    model: "o4-mini",
    useResponsesApi: true,
});

const llmWithMcp = llm.bindTools([
    {
        type: "mcp",
        server_label: "deepwiki",
        server_url: "https://mcp.deepwiki.com/mcp",
        require_approval: "never"
    }
]);

const response = await llmWithMcp.invoke(
    "What transport protocols does the 2025-03-26 version of the MCP spec (modelcontextprotocol/modelcontextprotocol) support?"
);

MCP 批准

当被指示时，OpenAI 将在对远程 MCP 服务器进行调用之前请求批准。

在上述命令中，我们指示模型从不要求批准。我们还可以配置模型始终请求批准，或始终针对特定工具请求批准：

...
const llmWithMcp = llm.bindTools([
  {
    type: "mcp",
    server_label: "deepwiki",
    server_url: "https://mcp.deepwiki.com/mcp",
    require_approval: {
      always: {
        tool_names: ["read_wiki_structure"],
      },
    },
  },
]);
const response = await llmWithMcp.invoke(
    "What transport protocols does the 2025-03-26 version of the MCP spec (modelcontextprotocol/modelcontextprotocol) support?"
);

使用此配置，响应可以包含类型为 mcp_approval_request 的工具输出。要提交批准请求的批准，您可以将其结构化为后续消息中的内容块：

const approvals = [];
if (Array.isArray(response.additional_kwargs.tool_outputs)) {
  for (const content of response.additional_kwargs.tool_outputs) {
    if (content.type === "mcp_approval_request") {
      approvals.push({
        type: "mcp_approval_response",
        approval_request_id: content.id,
        approve: true,
      });
    }
  }
}

const nextResponse = await model.invoke(
  [
    response,
    new HumanMessage({ content: approvals }),
  ],
);

图像生成

ChatOpenAI 允许您将内置的 图像生成工具 带入多轮对话中，通过响应 API 创建图像。

import { ChatOpenAI } from "@langchain/openai";

const llm = new ChatOpenAI({
  model: "gpt-4.1",
  useResponsesApi: true,
});

const llmWithImageGeneration = llm.bindTools([
  {
    type: "image_generation",
    quality: "low",
  }
]);

const response = await llmWithImageGeneration.invoke(
  "Draw a random short word in green font."
)

推理模型

兼容性：以下要点适用于 @langchain/openai>=0.4.0。

使用推理模型（如 o1）时，withStructuredOutput 的默认方法是 OpenAI 的结构化输出内置方法（相当于向 withStructuredOutput 传递选项 method: "jsonSchema"）。JSON schema 与其他模型基本相同，但有一个重要的注意事项：定义架构时，z.optional() 不被尊重，您应该改用 z.nullable()。

这是一个示例：

import * as z from "zod";
import { ChatOpenAI } from "@langchain/openai";

// Will not work
const reasoningModelSchemaOptional = z.object({
  color: z.optional(z.string()).describe("A color mentioned in the input"),
});

const reasoningModelOptionalSchema = new ChatOpenAI({
  model: "o1",
}).withStructuredOutput(reasoningModelSchemaOptional, {
  name: "extract_color",
});

await reasoningModelOptionalSchema.invoke([{
  role: "user",
  content: `I am 6'5" tall and love fruit.`
}]);

{ color: 'No color mentioned' }

这里是一个使用 z.nullable() 的示例：

import * as z from "zod";
import { ChatOpenAI } from "@langchain/openai";

// Will not work
const reasoningModelSchemaNullable = z.object({
  color: z.nullable(z.string()).describe("A color mentioned in the input"),
});

const reasoningModelNullableSchema = new ChatOpenAI({
  model: "o1",
}).withStructuredOutput(reasoningModelSchemaNullable, {
  name: "extract_color",
});

await reasoningModelNullableSchema.invoke([{
  role: "user",
  content: `I am 6'5" tall and love fruit.`
}]);

{ color: null }

提示词缓存

较新的 OpenAI 模型会自动 缓存您提示词的一部分，如果您的输入超过一定大小（撰写时为 1024 个 token），以减少需要长上下文的用例的成本。

注意： 给定查询缓存的 token 数量尚未在 AIMessage.usage_metadata 中标准化，而是包含在 AIMessage.response_metadata 字段中。

这是一个示例

// @lc-docs-hide-cell

const CACHED_TEXT = `## Components

LangChain provides standard, extendable interfaces and external integrations for various components useful for building with LLMs.
Some components LangChain implements, some components we rely on third-party integrations for, and others are a mix.

### Chat models

<span data-heading-keywords="chat model,chat models"></span>

Language models that use a sequence of messages as inputs and return chat messages as outputs (as opposed to using plain text).
These are generally newer models (older models are generally \`LLMs\`, see below).
Chat models support the assignment of distinct roles to conversation messages, helping to distinguish messages from the AI, users, and instructions such as system messages.

Although the underlying models are messages in, message out, the LangChain wrappers also allow these models to take a string as input.
This gives them the same interface as LLMs (and simpler to use).
When a string is passed in as input, it will be converted to a \`HumanMessage\` under the hood before being passed to the underlying model.

LangChain does not host any Chat Models, rather we rely on third party integrations.

We have some standardized parameters when constructing ChatModels:

- \`model\`: the name of the model

Chat Models also accept other parameters that are specific to that integration.

<Warning>
**Some chat models have been fine-tuned for **tool calling** and provide a dedicated API for it.**

Generally, such models are better at tool calling than non-fine-tuned models, and are recommended for use cases that require tool calling.
Please see the [tool calling section](/oss/javascript/langchain/tools) for more information.
</Warning>

For specifics on how to use chat models, see the [relevant how-to guides here](/oss/javascript/langchain/models).

#### Multimodality

Some chat models are multimodal, accepting images, audio and even video as inputs.
These are still less common, meaning model providers haven't standardized on the "best" way to define the API.
Multimodal outputs are even less common. As such, we've kept our multimodal abstractions fairly light weight
and plan to further solidify the multimodal APIs and interaction patterns as the field matures.

In LangChain, most chat models that support multimodal inputs also accept those values in OpenAI's content blocks format.
So far this is restricted to image inputs. For models like Gemini which support video and other bytes input, the APIs also support the native, model-specific representations.

For specifics on how to use multimodal models, see the [relevant how-to guides here](/oss/javascript/how-to/#multimodal).

### LLMs

<span data-heading-keywords="llm,llms"></span>

<Warning>
**Pure text-in/text-out LLMs tend to be older or lower-level. Many popular models are best used as [chat completion models](/oss/javascript/langchain/models),**

even for non-chat use cases.

You are probably looking for [the section above instead](/oss/javascript/langchain/models).
</Warning>

Language models that takes a string as input and returns a string.
These are traditionally older models (newer models generally are [Chat Models](/oss/javascript/langchain/models), see above).

Although the underlying models are string in, string out, the LangChain wrappers also allow these models to take messages as input.
This gives them the same interface as [Chat Models](/oss/javascript/langchain/models).
When messages are passed in as input, they will be formatted into a string under the hood before being passed to the underlying model.

LangChain does not host any LLMs, rather we rely on third party integrations.

For specifics on how to use LLMs, see the [relevant how-to guides here](/oss/javascript/langchain/models).

### Message types

Some language models take an array of messages as input and return a message.
There are a few different types of messages.
All messages have a \`role\`, \`content\`, and \`response_metadata\` property.

The \`role\` describes WHO is saying the message.
LangChain has different message classes for different roles.

The \`content\` property describes the content of the message.
This can be a few different things:

- A string (most models deal this type of content)
- A List of objects (this is used for multi-modal input, where the object contains information about that input type and that input location)

#### HumanMessage

This represents a message from the user.

#### AIMessage

This represents a message from the model. In addition to the \`content\` property, these messages also have:

**\`response_metadata\`**

The \`response_metadata\` property contains additional metadata about the response. The data here is often specific to each model provider.
This is where information like log-probs and token usage may be stored.

**\`tool_calls\`**

These represent a decision from an language model to call a tool. They are included as part of an \`AIMessage\` output.
They can be accessed from there with the \`.tool_calls\` property.

This property returns a list of \`ToolCall\`s. A \`ToolCall\` is an object with the following arguments:

- \`name\`: The name of the tool that should be called.
- \`args\`: The arguments to that tool.
- \`id\`: The id of that tool call.

#### SystemMessage

This represents a system message, which tells the model how to behave. Not every model provider supports this.

#### ToolMessage

This represents the result of a tool call. In addition to \`role\` and \`content\`, this message has:

- a \`tool_call_id\` field which conveys the id of the call to the tool that was called to produce this result.
- an \`artifact\` field which can be used to pass along arbitrary artifacts of the tool execution which are useful to track but which should not be sent to the model.

#### (Legacy) FunctionMessage

This is a legacy message type, corresponding to OpenAI's legacy function-calling API. \`ToolMessage\` should be used instead to correspond to the updated tool-calling API.

This represents the result of a function call. In addition to \`role\` and \`content\`, this message has a \`name\` parameter which conveys the name of the function that was called to produce this result.

### Prompt templates

<span data-heading-keywords="prompt,prompttemplate,chatprompttemplate"></span>

Prompt templates help to translate user input and parameters into instructions for a language model.
This can be used to guide a model's response, helping it understand the context and generate relevant and coherent language-based output.

Prompt Templates take as input an object, where each key represents a variable in the prompt template to fill in.

Prompt Templates output a PromptValue. This PromptValue can be passed to an LLM or a ChatModel, and can also be cast to a string or an array of messages.
The reason this PromptValue exists is to make it easy to switch between strings and messages.

There are a few different types of prompt templates:

#### String PromptTemplates

These prompt templates are used to format a single string, and generally are used for simpler inputs.
For example, a common way to construct and use a PromptTemplate is as follows:

\`\`\`typescript
import { PromptTemplate } from "@langchain/core/prompts";

const promptTemplate = PromptTemplate.fromTemplate(
  "Tell me a joke about {topic}"
);

await promptTemplate.invoke({ topic: "cats" });
\`\`\`

#### ChatPromptTemplates

These prompt templates are used to format an array of messages. These "templates" consist of an array of templates themselves.
For example, a common way to construct and use a ChatPromptTemplate is as follows:

\`\`\`typescript
import { ChatPromptTemplate } from "@langchain/core/prompts";

const promptTemplate = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant"],
  ["user", "Tell me a joke about {topic}"],
]);

await promptTemplate.invoke({ topic: "cats" });
\`\`\`

In the above example, this ChatPromptTemplate will construct two messages when called.
The first is a system message, that has no variables to format.
The second is a HumanMessage, and will be formatted by the \`topic\` variable the user passes in.

#### MessagesPlaceholder

<span data-heading-keywords="messagesplaceholder"></span>

This prompt template is responsible for adding an array of messages in a particular place.
In the above ChatPromptTemplate, we saw how we could format two messages, each one a string.
But what if we wanted the user to pass in an array of messages that we would slot into a particular spot?
This is how you use MessagesPlaceholder.

\`\`\`typescript
import {
  ChatPromptTemplate,
  MessagesPlaceholder,
} from "@langchain/core/prompts";
import { HumanMessage } from "@langchain/core/messages";

const promptTemplate = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant"],
  new MessagesPlaceholder("msgs"),
]);

promptTemplate.invoke({ msgs: [new HumanMessage({ content: "hi!" })] });
\`\`\`

This will produce an array of two messages, the first one being a system message, and the second one being the HumanMessage we passed in.
If we had passed in 5 messages, then it would have produced 6 messages in total (the system message plus the 5 passed in).
This is useful for letting an array of messages be slotted into a particular spot.

An alternative way to accomplish the same thing without using the \`MessagesPlaceholder\` class explicitly is:

\`\`\`typescript
const promptTemplate = ChatPromptTemplate.fromMessages([
  ["system", "You are a helpful assistant"],
  ["placeholder", "{msgs}"], // <-- This is the changed part
]);
\`\`\`

For specifics on how to use prompt templates, see the [relevant how-to guides here](/oss/javascript/how-to/#prompt-templates).

### Example Selectors

One common prompting technique for achieving better performance is to include examples as part of the prompt.
This gives the language model concrete examples of how it should behave.
Sometimes these examples are hardcoded into the prompt, but for more advanced situations it may be nice to dynamically select them.
Example Selectors are classes responsible for selecting and then formatting examples into prompts.

For specifics on how to use example selectors, see the [relevant how-to guides here](/oss/javascript/how-to/#example-selectors).

### Output parsers

<span data-heading-keywords="output parser"></span>

<Note>
**The information here refers to parsers that take a text output from a model try to parse it into a more structured representation.**

More and more models are supporting function (or tool) calling, which handles this automatically.
It is recommended to use function/tool calling rather than output parsing.
See the [LangChain tools documentation](/oss/javascript/langchain/tools).

</Note>

Responsible for taking the output of a model and transforming it to a more suitable format for downstream tasks.
Useful when you are using LLMs to generate structured data, or to normalize output from chat models and LLMs.

There are two main methods an output parser must implement:

- "Get format instructions": A method which returns a string containing instructions for how the output of a language model should be formatted.
- "Parse": A method which takes in a string (assumed to be the response from a language model) and parses it into some structure.

And then one optional one:

- "Parse with prompt": A method which takes in a string (assumed to be the response from a language model) and a prompt (assumed to be the prompt that generated such a response) and parses it into some structure. The prompt is largely provided in the event the OutputParser wants to retry or fix the output in some way, and needs information from the prompt to do so.

Output parsers accept a string or \`BaseMessage\` as input and can return an arbitrary type.

LangChain has many different types of output parsers. This is a list of output parsers LangChain supports. The table below has various pieces of information:

**Name**: The name of the output parser

**Supports Streaming**: Whether the output parser supports streaming.

**Input Type**: Expected input type. Most output parsers work on both strings and messages, but some (like OpenAI Functions) need a message with specific arguments.

**Output Type**: The output type of the object returned by the parser.

**Description**: Our commentary on this output parser and when to use it.

The current date is ${new Date().toISOString()}`;

// Noop statement to hide output
void 0;

import { ChatOpenAI } from "@langchain/openai";

const modelWithCaching = new ChatOpenAI({
  model: "gpt-4.1-mini-2025-04-14",
});

// CACHED_TEXT is some string longer than 1024 tokens
const LONG_TEXT = `You are a pirate. Always respond in pirate dialect.

Use the following as context when answering questions:

${CACHED_TEXT}`;

const longMessages = [
  {
    role: "system",
    content: LONG_TEXT,
  },
  {
    role: "user",
    content: "What types of messages are supported in LangChain?",
  },
];

const originalRes = await modelWithCaching.invoke(longMessages);

console.log("USAGE:", originalRes.response_metadata.usage);

USAGE: {
  prompt_tokens: 2624,
  completion_tokens: 263,
  total_tokens: 2887,
  prompt_tokens_details: { cached_tokens: 0 },
  completion_tokens_details: { reasoning_tokens: 0 }
}

const resWitCaching = await modelWithCaching.invoke(longMessages);

console.log("USAGE:", resWitCaching.response_metadata.usage);

USAGE: {
  prompt_tokens: 2624,
  completion_tokens: 272,
  total_tokens: 2896,
  prompt_tokens_details: { cached_tokens: 2432 },
  completion_tokens_details: { reasoning_tokens: 0 }
}

预测输出

一些 OpenAI 模型（如它们的 gpt-4o 和 gpt-4o-mini 系列）支持 预测输出，允许您提前传递 LLM 预期输出的一部分以减少延迟。这对于编辑文本或代码等情况很有用，因为模型的输出只有很小一部分会改变。

这是一个示例：

import { ChatOpenAI } from "@langchain/openai";

const modelWithPredictions = new ChatOpenAI({
  model: "gpt-4.1-mini",
});

const codeSample = `
/// <summary>
/// Represents a user with a first name, last name, and username.
/// </summary>
public class User
{
/// <summary>
/// Gets or sets the user's first name.
/// </summary>
public string FirstName { get; set; }

/// <summary>
/// Gets or sets the user's last name.
/// </summary>
public string LastName { get; set; }

/// <summary>
/// Gets or sets the user's username.
/// </summary>
public string Username { get; set; }
}
`;

// Can also be attached ahead of time
// using `model.bind({ prediction: {...} })`;
await modelWithPredictions.invoke(
  [
    {
      role: "user",
      content:
        "Replace the Username property with an Email property. Respond only with code, and with no markdown formatting.",
    },
    {
      role: "user",
      content: codeSample,
    },
  ],
  {
    prediction: {
      type: "content",
      content: codeSample,
    },
  }
);

AIMessage {
  "id": "chatcmpl-AQLyQKnazr7lEV7ejLTo1UqhzHDBl",
  "content": "/// <summary>\n/// Represents a user with a first name, last name, and email.\n/// </summary>\npublic class User\n{\n/// <summary>\n/// Gets or sets the user's first name.\n/// </summary>\npublic string FirstName { get; set; }\n\n/// <summary>\n/// Gets or sets the user's last name.\n/// </summary>\npublic string LastName { get; set; }\n\n/// <summary>\n/// Gets or sets the user's email.\n/// </summary>\npublic string Email { get; set; }\n}",
  "additional_kwargs": {},
  "response_metadata": {
    "tokenUsage": {
      "promptTokens": 148,
      "completionTokens": 217,
      "totalTokens": 365
    },
    "finish_reason": "stop",
    "usage": {
      "prompt_tokens": 148,
      "completion_tokens": 217,
      "total_tokens": 365,
      "prompt_tokens_details": {
        "cached_tokens": 0
      },
      "completion_tokens_details": {
        "reasoning_tokens": 0,
        "accepted_prediction_tokens": 36,
        "rejected_prediction_tokens": 116
      }
    },
    "system_fingerprint": "fp_0ba0d124f1"
  },
  "tool_calls": [],
  "invalid_tool_calls": [],
  "usage_metadata": {
    "output_tokens": 217,
    "input_tokens": 148,
    "total_tokens": 365,
    "input_token_details": {
      "cache_read": 0
    },
    "output_token_details": {
      "reasoning": 0
    }
  }
}

注意，目前预测被视为额外 token 计费，并将增加您的使用量和成本，以换取这种降低的延迟。

音频输出

一些 OpenAI 模型（如 gpt-4o-audio-preview）支持生成音频输出。此示例展示了如何使用该功能：

import { ChatOpenAI } from "@langchain/openai";

const modelWithAudioOutput = new ChatOpenAI({
  model: "gpt-4o-audio-preview",
  // You may also pass these fields to `.bind` as a call argument.
  modalities: ["text", "audio"], // Specifies that the model should output audio.
  audio: {
    voice: "alloy",
    format: "wav",
  },
});

const audioOutputResult = await modelWithAudioOutput.invoke("Tell me a joke about cats.");
const castAudioContent = audioOutputResult.additional_kwargs.audio as Record<string, any>;

console.log({
  ...castAudioContent,
  data: castAudioContent.data.slice(0, 100) // Sliced for brevity
})

{
  id: 'audio_67129e9466f48190be70372922464162',
  data: 'UklGRgZ4BABXQVZFZm10IBAAAAABAAEAwF0AAIC7AAACABAATElTVBoAAABJTkZPSVNGVA4AAABMYXZmNTguMjkuMTAwAGRhdGHA',
  expires_at: 1729277092,
  transcript: "Why did the cat sit on the computer's keyboard? Because it wanted to keep an eye on the mouse!"
}

我们看到音频数据在 data 字段内返回。我们还提供了一个 expires_at 日期字段。此字段表示音频响应不再可在服务器上访问的日期，以便在多轮对话中使用。

流式音频输出

OpenAI 还支持流式音频输出。这是一个示例：

import { AIMessageChunk } from "@langchain/core/messages";
import { concat } from "@langchain/core/utils/stream"
import { ChatOpenAI } from "@langchain/openai";

const modelWithStreamingAudioOutput = new ChatOpenAI({
  model: "gpt-4o-audio-preview",
  modalities: ["text", "audio"],
  audio: {
    voice: "alloy",
    format: "pcm16", // Format must be `pcm16` for streaming
  },
});

const audioOutputStream = await modelWithStreamingAudioOutput.stream("Tell me a joke about cats.");
let finalAudioOutputMsg: AIMessageChunk | undefined;
for await (const chunk of audioOutputStream) {
  finalAudioOutputMsg = finalAudioOutputMsg ? concat(finalAudioOutputMsg, chunk) : chunk;
}
const castStreamedAudioContent = finalAudioOutputMsg?.additional_kwargs.audio as Record<string, any>;

console.log({
  ...castStreamedAudioContent,
  data: castStreamedAudioContent.data.slice(0, 100) // Sliced for brevity
})

{
  id: 'audio_67129e976ce081908103ba4947399a3eaudio_67129e976ce081908103ba4947399a3e',
  transcript: 'Why was the cat sitting on the computer? Because it wanted to keep an eye on the mouse!',
  index: 0,
  data: 'CgAGAAIADAAAAA0AAwAJAAcACQAJAAQABQABAAgABQAPAAAACAADAAUAAwD8/wUA+f8MAPv/CAD7/wUA///8/wUA/f8DAPj/AgD6',
  expires_at: 1729277096
}

音频输入

这些模型还支持传递音频作为输入。为此，您必须指定 input_audio 字段，如下所示：

import { HumanMessage } from "@langchain/core/messages";

const userInput = new HumanMessage({
  content: [{
    type: "input_audio",
    input_audio: {
      data: castAudioContent.data, // Reuse the base64 data from the first example
      format: "wav",
    },
  }]
})

// Reuse the same model instance
const userInputAudioRes = await modelWithAudioOutput.invoke([userInput]);

console.log((userInputAudioRes.additional_kwargs.audio as Record<string, any>).transcript);

That's a great joke! It's always fun to imagine why cats do the funny things they do. Keeping an eye on the "mouse" is a creatively punny way to describe it!

---

API 参考

有关所有 ChatOpenAI 功能和配置的详细文档，请前往 API 参考。

---

Edit this page on GitHub or file an issue.

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.