Langchain V2

python.langchain.com/docs covers LCEL guide, chain patterns, agent frameworks, memory, RAG patterns, and 100+ integration docs. Getting started: pip install langchain langchain-openai — first chain in 3 minutes. Fully free and open-source (MIT). GitHub: langchain-ai/langchain (100k+ stars). Community via Discord (very active, 100k+ members). Comprehensive but rapidly evolving documentation; LangChain has the most integrations and examples of any LLM framework.

LangChain is the most widely adopted LLM framework — 100k+ GitHub stars, the largest ecosystem of integrations, and the de facto standard for wiring LLM components into applications. For agents: build RAG pipelines with LCEL (retriever | prompt | llm | StrOutputParser()), use pre-built chains (ConversationalRetrievalChain, MapReduceChain), build tool-using agents (create_tool_calling_agent + AgentExecutor), and manage multi-turn conversation memory. LangGraph extends LangChain with stateful, cyclical multi-agent graphs. LangSmith auto-traces LangChain runs. Confidence is docs-derived.

Python SDK: pip install langchain langchain-openai. LCEL chain: from langchain_openai import ChatOpenAI; from langchain_core.prompts import ChatPromptTemplate; from langchain_core.output_parsers import StrOutputParser; model = ChatOpenAI(model="gpt-4o"); prompt = ChatPromptTemplate.from_template("Answer: {question}"); chain = prompt | model | StrOutputParser(); result = chain.invoke({"question": "What is 2+2?"}). RAG chain: from langchain.chains import create_retrieval_chain; from langchain.chains.combine_documents import create_stuff_documents_chain; combine_docs_chain = create_stuff_documents_chain(model, prompt); retrieval_chain = create_retrieval_chain(retriever, combine_docs_chain); result = retrieval_chain.invoke({"input": "user question"}). Tool agent: from langchain.agents import create_tool_calling_agent, AgentExecutor; agent = create_tool_calling_agent(model, tools, prompt); executor = AgentExecutor(agent=agent, tools=tools); executor.invoke({"input": "search for X"}). Streaming: for chunk in chain.stream({"question": "..."}): print(chunk, end="", flush=True).

No LangChain API auth — pure local library. LLM provider keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) read from env vars by respective provider packages. LangSmith tracing: LANGCHAIN_API_KEY + LANGCHAIN_TRACING_V2=true. Vector store auth: passed to vector store client. No OAuth2 natively. LangChain Hub (prompt registry): LangChain API key for pulling/pushing prompts.

LCEL: lazy evaluation — chains are composable runnables; only executed on invoke/stream/batch. Streaming: built-in with .stream(); works across all LCEL components. Async: .ainvoke()/.astream() for async/await. Token counting: callbacks for token tracking (get_openai_callback()). Callbacks: configurable for logging, tracing, and streaming. Memory: BaseMemory implementations for conversation history (ConversationBufferMemory, ConversationSummaryMemory). LangGraph: stateful graph execution with persistence backends (SQLite, PostgreSQL). Deprecation cadence: LangChain evolves rapidly — pin versions in production.