langchain

Use create_agent with a chat model, a list of tool functions, and a system prompt. Define tools as Python functions with docstrings describing their purpose, then invoke the agent with a messages list containing the user query.

Load documents with a loader like WebBaseLoader, split them with RecursiveCharacterTextSplitter, embed them into a vector store such as Chroma, and connect a retriever to a RetrievalQA chain. The chain retrieves relevant chunks and generates answers with source documents.

LangChain suits quick agents and RAG with high-level abstractions and minimal code. LangGraph fits complex stateful workflows needing cycles, fine-grained control, multi-agent systems, and human-in-the-loop patterns.

Yes, LangChain supports OpenAI, Anthropic, Google, Azure OpenAI, and local models through Ollama. Swap providers by changing the chat model class, such as ChatOpenAI to ChatAnthropic, without rewriting application logic.

Use 512 to 1024 tokens for general question answering, with 10 to 20 percent overlap between chunks. Shorter chunks of 256 to 512 tokens suit short answers, while 1024 to 2048 tokens work for long-context retrieval.

Common causes include vague tool descriptions, too many tools, missing error handling, and no iteration limit. Set max_iterations on the AgentExecutor, add try-except blocks in tools, and enable verbose mode to inspect reasoning steps.