langchain

par langchain-ai

Construisez des agents avec une architecture préconstruite et des intégrations pour tout modèle ou outil. Utilisez lors de la création d'agents appelant des outils, du changement de fournisseurs de modèles, ou de l'ajout…

npx skills add https://github.com/langchain-ai/docs --skill langchain

Télécharger ZIP GitHub

421

LangChain

LangChain is an open-source framework with a prebuilt agent architecture and integrations for any model or tool. Build agents and LLM-powered applications in under 10 lines of code, with integrations for OpenAI, Anthropic, Google, and hundreds more.

When to use

Use LangChain when you need to:

Build tool-calling agents with create_agent() and a prebuilt agent loop
Switch model providers without changing application code via init_chat_model()
Add structured output to parse LLM responses into typed objects
Integrate with any model or tool using LangChain's provider packages
Use middleware for cross-cutting concerns like rate limiting and caching

When NOT to use

For complex multi-step workflows with custom control flow, use LangGraph instead
For a batteries-included agent with planning, subagents, and context management, use Deep Agents instead
LangChain provides the core building blocks; LangGraph adds orchestration; Deep Agents adds high-level capabilities on top

Install

# Python
pip install -U langchain

# JavaScript/TypeScript
npm install langchain @langchain/core

Install a provider integration:

# Python
pip install -U langchain-openai       # or langchain-anthropic, langchain-google-genai

# JavaScript/TypeScript
npm install @langchain/openai         # or @langchain/anthropic, @langchain/google-genai

Quick reference

Create an agent

from langchain.agents import create_agent

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

agent = create_agent(
    model="openai:gpt-5.5",
    tools=[get_weather],
    system_prompt="You are a helpful assistant",
)

result = agent.invoke(
    {"messages": [{"role": "user", "content": "What is the weather in SF?"}]}
)

Initialize a chat model

from langchain.chat_models import init_chat_model

# Switch providers by changing the string
model = init_chat_model("openai:gpt-5.5")
model = init_chat_model("anthropic:claude-opus-4-8")
model = init_chat_model("google_genai:gemini-3.6-flash")

Define a tool

from langchain.tools import tool

@tool
def search(query: str) -> str:
    """Search the web for information."""
    return "search results"

Gotchas

Snake_case tool names—Tool function names must be valid Python identifiers. Use get_weather, not get-weather.
Reserved parameters—Do not name tool parameters type, name, or description as these conflict with the tool schema.
Provider packages—Models live in separate packages (e.g., langchain-openai). The base langchain package does not include providers.
Model string format—Use "provider:model-name" format with init_chat_model() (e.g., "openai:gpt-5.5").