LangChain Ecosystem

LangChain looks like one tool from the outside. Under the surface, it works as a family of small packages and services, and each member handles one job. Picture a city: roads, water pipes, and traffic lights belong to different departments, yet residents experience one smooth city. The LangChain ecosystem follows the same idea, so you install only the parts your project needs.

The Big Picture

Three products sit at the top of the family. Each one solves a different stage of building an AI application.

 BUILD                 ORCHESTRATE             OBSERVE
 langchain             langgraph               langsmith
 (agents, chains)  →   (workflows, loops)  →   (tracing, testing)
        \                    |                     /
         \                   |                    /
          ------- langchain-core ----------------
                   (shared building blocks)
                          |
        ------------------------------------------
        |              |              |          |
   langchain-openai  langchain-anthropic  langchain-google-genai  ...
        (one small package per model provider)

langchain-core: The Foundation

The core package defines the shared vocabulary: messages, prompts, parsers, runnables, and tools. Every other package builds on it. When two packages speak the same vocabulary, you can swap a model from one company for a model from another without rewriting your chain.

langchain: Ready-Made Patterns

The main package gives you high-level helpers. The most popular helper creates an agent in a few lines. It also holds the standard way to start any chat model from a simple text name.

Provider Packages

Each model company ships its own integration package. OpenAI users install langchain-openai. Anthropic users install langchain-anthropic. Google users install langchain-google-genai. Small packages keep your project light, because you never download code for providers you skip.

langchain-community

The community package collects thousands of extra integrations contributed by developers worldwide. Document loaders for rare file types and connectors for lesser-known databases live here. Search the community package when a popular package lacks the connector you need.

LangGraph: Workflows with Loops

A simple chain runs from start to finish in a straight line. Real applications often need to loop, branch, pause, or ask a human for approval. LangGraph handles these cases by treating your app as a map of steps connected by arrows. Topics 32 to 36 explore LangGraph in depth.

LangSmith: Your Application Dashboard

LangSmith records every step your application takes. You can see the exact prompt, the model reply, the time spent, and the cost. Think of it as a flight recorder for your AI app. When an answer looks wrong, the recording shows where things went off track.

One Application, Many Packages

A small support bot shows how the pieces cooperate.

 User question
      |
      v
 langchain-openai   → talks to the chat model
 langchain-core     → formats the prompt and parses the reply
 langchain          → creates the agent
 langgraph          → runs the agent loop under the hood
 langsmith          → records the whole trip
      |
      v
 Final answer

Installing Only What You Need

pip install langchain langchain-openai
pip install langgraph
pip install langsmith
pip install langchain-community

Start with the first line. Add the others when a project demands them.

Beginner Tips

  • Read import lines carefully. The package name in the import tells you where a feature lives.
  • Install one provider package at a time to keep your environment clean.
  • Check the version of each package when a tutorial example fails. Fast-moving libraries change names between releases.
  • Create a separate virtual environment for each project to avoid version clashes.

Real-World Scenario

A freelance developer builds a booking assistant for a small hotel. The developer installs langchain and langchain-openai for the first version. Guests ask questions about rooms and prices, and the assistant answers. Two months later, the hotel wants approval steps for refunds, so the developer adds langgraph. A tracing dashboard from langsmith helps the developer find slow replies before guests notice. The project grew one package at a time, and each addition solved a real need.

Practice Tasks

  1. Draw your own ecosystem map for an app that answers questions about school timetables.
  2. List which package you would install first and which you would add later.
  3. Create a fresh virtual environment and install langchain-core alone. Run a small import to confirm it works.
  4. Compare the install size of a project with one provider package against a project with three.

Key Points

  • LangChain is a family of packages, not a single library.
  • langchain-core holds the shared building blocks.
  • Provider packages connect your code to model companies.
  • LangGraph builds workflows with loops and branches.
  • LangSmith traces, tests, and monitors your application.
Previous lesson
Back to course
Next lesson