1. 简介

LLM 大模型应用开发框架, 用来构建 agent.
LangChain 提供统一接口，可连接：大模型、Prompt、向量数据库、工具调用、记忆系统以及 Agent 工作流.
它是 python 语言, 虽然也有 js 版本叫 LangChain.js, 但生态和更新速度都不如 Python.

2. 核心组件

LLM：连接 OpenAI、Claude、Gemini 等大模型
PromptTemplate：管理 Prompt 模板
Chains：构建多步骤 AI 工作流
Memory：实现多轮对话记忆
Tools：调用搜索、数据库、API 等工具
Agents：让 AI 自动决策与执行任务
Vector Store：连接向量数据库实现 RAG

3. 极简 demo

最简单的 LangChain + 阿里云大模型 Demo

import sys
sys.stdout.reconfigure(encoding="utf-8")

from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompt_values import ChatPromptValue
from langchain_core.messages import AIMessage

load_dotenv()

llm = ChatOpenAI(model="qwen-plus")

def demo_lcel():
    """写法一：LCEL 管道风格（推荐）"""
    chain = (
        ChatPromptTemplate.from_messages([("human", "{question}")])
        | llm
        | StrOutputParser()
    )

    result = chain.invoke({"question": "你好，请用一句话介绍你自己"})
    print(result)


def demo_step_by_step() -> None:
    """写法二：普通逐步调用"""
    prompt: ChatPromptTemplate = ChatPromptTemplate.from_messages([("human", "{question}")])
    parser: StrOutputParser = StrOutputParser()

    # 第一步：模板填充变量，生成提示词
    prompt_value: ChatPromptValue = prompt.invoke({"question": "你好，请用一句话介绍你自己"})

    # 第二步：调用 LLM，返回 AIMessage 对象
    ai_message: AIMessage = llm.invoke(prompt_value)

    # 第三步：解析器提取纯文本
    result: str = parser.invoke(ai_message)
    print(result)


# demo_lcel()
demo_step_by_step()

4. llm.invoke() 调用链

4.1 http 请求

背后是 httpx 框架的 http post 请求.

ChatOpenAI.invoke()                    ← langchain-core 入口
  └→ _generate()                       ← langchain_openai/chat_models/base.py:1695
      └→ self.client.with_raw_response.create(**payload)   ← 进入 openai SDK
          └→ SyncAPIClient.request()   ← openai/_base_client.py:995
              └→ _send_request()       ← openai/_base_client.py:957
                  └→ self._client.send(request)            ← httpx.Client.send() 🔥 真正的 HTTP 请求

4.2 LLM 传参

调用 LLM 的相关参数见下.

# 位于 site-packages/openai/resources/chat/completions/completions.py
class Completions(SyncAPIResource):
    @required_args(["messages", "model"], ["messages", "model", "stream"])
    def create(...):
        return self._post(
                    "/chat/completions",
                    body=maybe_transform(
                        {
                            "messages": messages,
                            "model": model,
                            "audio": audio,
                            "frequency_penalty": frequency_penalty,
                            "function_call": function_call,
                            "functions": functions,
                            "logit_bias": logit_bias,
                            "logprobs": logprobs,
                            "max_completion_tokens": max_completion_tokens,
                            "max_tokens": max_tokens,
                            "metadata": metadata,
                            "modalities": modalities,
                            "moderation": moderation,
                            "n": n,
                            "parallel_tool_calls": parallel_tool_calls,
                            "prediction": prediction,
                            "presence_penalty": presence_penalty,
                            "prompt_cache_key": prompt_cache_key,
                            "prompt_cache_options": prompt_cache_options,
                            "prompt_cache_retention": prompt_cache_retention,
                            "reasoning_effort": reasoning_effort,
                            "response_format": response_format,
                            "safety_identifier": safety_identifier,
                            "seed": seed,
                            "service_tier": service_tier,
                            "stop": stop,
                            "store": store,
                            "stream": stream,
                            "stream_options": stream_options,
                            "temperature": temperature,
                            "tool_choice": tool_choice,
                            "tools": tools,
                            "top_logprobs": top_logprobs,
                            "top_p": top_p,
                            "user": user,
                            "verbosity": verbosity,
                            "web_search_options": web_search_options,
                        },
                        completion_create_params.CompletionCreateParamsStreaming
                        if stream
                        else completion_create_params.CompletionCreateParamsNonStreaming,
                    ),
                    options=make_request_options(
                        extra_headers=extra_headers,
                        extra_query=extra_query,
                        extra_body=extra_body,
                        timeout=timeout,
                        security={"bearer_auth": True},
                    ),
                    cast_to=ChatCompletion,
                    stream=stream or False,
                    stream_cls=Stream[ChatCompletionChunk],
                )

附录

1. LangChain 官网