Langchain Use

LangChain Use Skill

LangChain 是构建 LLM 驱动的智能体和应用程序的开源框架。

安装

使用 uv 安装 LangChain（推荐，需要 Python 3.10+）：

# 安装核心包
uv add langchain

# 安装模型提供商集成
uv add langchain-anthropic  # Anthropic/Claude
uv add langchain-openai     # OpenAI

快速参考

核心工作流程

用户查询 -> create_agent() -> ReAct 循环 -> Tool 调用 -> 返回结果

创建 Agent

详见 Agent 基础

from langchain.agents import create_agent

agent = create_agent(
    model="claude-sonnet-4-5-20250929",
    tools=[get_weather],
    system_prompt="You are a helpful assistant",
)

# 运行 agent
result = agent.invoke(
    {"messages": [{"role": "user", "content": "what is the weather in sf"}]}
)

定义 Tool

详见 Tool 基础

from langchain.tools import tool

@tool
def get_weather(city: str) -> str:
    """Get weather for a given city."""
    return f"It's always sunny in {city}!"

访问 Runtime Context

使用 ToolRuntime 访问 state、context、store：

from langchain.tools import tool, ToolRuntime
from dataclasses import dataclass

@dataclass
class Context:
    user_id: str

@tool
def get_user_location(runtime: ToolRuntime[Context]) -> str:
    """Retrieve user location based on user ID."""
    user_id = runtime.context.user_id
    return "Florida" if user_id == "1" else "SF"

管理 Memory

详见 短期记忆 详见 长期记忆

from langgraph.checkpoint.memory import InMemorySaver

agent = create_agent(
    model,
    tools,
    checkpointer=InMemorySaver(),  # 短期记忆
)

# 使用 thread_id 维护会话
config = {"configurable": {"thread_id": "1"}}
agent.invoke({"messages": [...]}, config)

添加 Middleware

详见 中间件概述

from langchain.agents.middleware import before_model, after_model

@before_model
def trim_messages(state, runtime):
    # 消息修剪逻辑
    return None

进阶主题

集成主题

关键概念

Agent (智能体)

LangChain 1.0 的核心抽象，基于 LangGraph 构建。使用 create_agent() 创建。

Tool (工具)

使用 @tool 装饰器定义。可选 ToolRuntime 访问状态、上下文和存储。

Memory (记忆)

• Checkpointer: 短期会话记忆 (InMemorySaver, PostgresSaver)
• Store: 长期持久化存储 (InMemoryStore)

Middleware (中间件)

装饰器风格的扩展机制：

• @before_model - 模型调用前处理
• @after_model - 模型调用后处理
• @wrap_tool_call - 工具调用包装
• @dynamic_prompt - 动态系统提示

常见模式

基础 Agent

from langchain.agents import create_agent

agent = create_agent(
    model="claude-sonnet-4-5-20250929",
    tools=[my_tool],
    system_prompt="You are helpful.",
)

result = agent.invoke(
    {"messages": [{"role": "user", "content": "hello"}]}
)

带记忆的 Agent（会话持久化）

from langgraph.checkpoint.memory import InMemorySaver

agent = create_agent(
    model="claude-sonnet-4-5-20250929",
    tools=[my_tool],
    checkpointer=InMemorySaver(),
)

# 使用 thread_id 标识会话
config = {"configurable": {"thread_id": "1"}}
agent.invoke(
    {"messages": [{"role": "user", "content": "My name is Bob"}]},
    config
)
agent.invoke(
    {"messages": [{"role": "user", "content": "What's my name?"}]},
    config
)

生产环境 Memory（使用数据库）

from langgraph.checkpoint.postgres import PostgresSaver

DB_URI = "postgresql://postgres:postgres@localhost:5432/postgres"
with PostgresSaver.from_conn_string(DB_URI) as checkpointer:
    checkpointer.setup()  # 自动创建表
    agent = create_agent(
        model="claude-sonnet-4-5-20250929",
        tools=[my_tool],
        checkpointer=checkpointer,
    )

带上下文的 Agent（Runtime Context）

from dataclasses import dataclass
from langchain.tools import tool, ToolRuntime

@dataclass
class Context:
    user_id: str

@tool
def get_user_info(runtime: ToolRuntime[Context]) -> str:
    """Get user information."""
    user_id = runtime.context.user_id
    return f"User ID: {user_id}"

agent = create_agent(
    model="claude-sonnet-4-5-20250929",
    tools=[get_user_info],
    context_schema=Context,
)

result = agent.invoke(
    {"messages": [{"role": "user", "content": "get my info"}]},
    context=Context(user_id="user_123")
)

带结构化输出的 Agent

from dataclasses import dataclass
from langchain.agents.structured_output import ToolStrategy

@dataclass
class Response:
    answer: str
    confidence: float

agent = create_agent(
    model="claude-sonnet-4-5-20250929",
    tools=[my_tool],
    response_format=ToolStrategy(Response),
)

result = agent.invoke(
    {"messages": [{"role": "user", "content": "what is 2+2?"}]}
)
print(result['structured_response'])
# Response(answer="4", confidence=1.0)

完整示例（生产级 Agent）

from dataclasses import dataclass
from langchain.agents import create_agent
from langchain.chat_models import init_chat_model
from langchain.tools import tool, ToolRuntime
from langgraph.checkpoint.memory import InMemorySaver
from langchain.agents.structured_output import ToolStrategy

@dataclass
class Context:
    user_id: str

@dataclass
class ResponseFormat:
    answer: str
    confidence: float | None = None

@tool
def get_weather(city: str) -> str:
    """Get weather for a city."""
    return f"Sunny in {city}"

@tool
def get_user_location(runtime: ToolRuntime[Context]) -> str:
    """Get user's location."""
    return "San Francisco" if runtime.context.user_id == "1" else "Unknown"

# 配置模型
model = init_chat_model(
    "claude-sonnet-4-5-20250929",
    temperature=0.5,
    max_tokens=1000
)

# 创建 agent
agent = create_agent(
    model=model,
    system_prompt="You are a weather assistant.",
    tools=[get_weather, get_user_location],
    context_schema=Context,
    response_format=ToolStrategy(ResponseFormat),
    checkpointer=InMemorySaver(),
)

# 运行 agent
config = {"configurable": {"thread_id": "1"}}
result = agent.invoke(
    {"messages": [{"role": "user", "content": "What's the weather?"}]},
    config=config,
    context=Context(user_id="1")
)

资源链接

• 核心概念 - LangChain 概述
• 快速开始 - 10行代码创建 Agent
• 官方文档
• API 参考