The LangChain ecosystem
"LangChain" is often used loosely to mean four separate projects that compose together: LangChain itself, LangGraph, LangSmith, and LangServe.
| Project | What it does | You need it when | Covered in |
|---|---|---|---|
| LangChain | Building blocks: chat models, prompts, retrieval, tools, output parsing | You're assembling any LLM-powered pipeline | This whole section |
| LangGraph | Stateful, cyclic orchestration of agents and multi-step workflows | Your flow loops, branches on model decisions, or needs to pause/resume | Why LangGraph |
| LangSmith | Tracing, evaluation, and prompt management | You need to see what a chain/agent actually did, or regression-test prompt changes | Tracing |
| LangServe | Exposing a Runnable as an HTTP API | You're serving a chain over HTTP and want schemas/streaming for free | LangServe |
How they relate at runtime
LangChain supplies the components. LangGraph decides the control flow when a single linear chain isn't enough. LangSmith watches both, regardless of which one is running. LangServe (or a hand-rolled FastAPI app) puts the result behind an HTTP endpoint.
tip
You don't need all four for every project. A single retrieval-augmented chain is LangChain alone. An agent that loops and calls tools usually reaches for LangGraph. Tracing is worth turning on early, even in development — see LangSmith tracing.