In plain words
LangChain is not an AI model or a chatbot. It is a free code library, a box of ready-made building blocks that programmers use when they build their own AI apps.
A useful AI app rarely just sends one message to a language model. It might search the company's documents, call a calculator or a booking system, check the answer and only then reply. LangChain gives developers ready parts for each of those steps and a way to chain them together, which is where the name comes from.
It also lets developers switch between models from different companies with little change to their code. LangChain is also the name of the company that makes it, which sells paid tools for testing and running AI agents.
Everyday example
A kitchen of standard parts
A modular kitchen comes as standard cabinets, a sink unit and a hob that fit together. You still decide the layout, but you do not build each cupboard from wood. LangChain gives programmers standard parts for AI apps: a model connector, a document search, a tool caller. They arrange them into their own design.
Where the example breaks: Kitchen parts behave the same every time. The AI model inside a LangChain app does not: the same steps can give different answers, so builders still have to test carefully.
How it actually works
LangChain has grown into a small family of open-source libraries plus a paid platform.
Bring in your own documents
Loaders read PDFs, web pages and databases, split them into chunks and store them as embeddings in a vector database. At question time the app finds the most relevant chunks and gives them to the model. This pattern is RAG.
Give the model tools
Developers describe functions, such as "look up an order" or "search the web", and the model decides when to call them. This is tool use, and a model that loops through thinking and tool calls until a task is done is an agent.
Control the steps with LangGraph
For complex agents, the sister library LangGraph lays out each step as a node in a graph, with memory, retries and pauses for a human to approve an action. Since version 1.0 in October 2025, LangChain's own agents run on LangGraph.
Watch and test with LangSmith
LangSmith, the company's paid platform, records every step an app takes so developers can see why an answer went wrong, score answers against tests and deploy agents.
- Started by
- Harrison Chase, as an open-source project in October 2022
- Licence
- MIT, free to use and change
- Sister libraries
- LangGraph for step-by-step agent control and Deep Agents for long-running tasks
- Version 1.0
- 22 October 2025, for both LangChain and LangGraph
LangChain is one of several agent frameworks; the open standard MCP is a separate way to connect models to tools that many frameworks support.
Questions people ask
Is LangChain an AI model?
No. It contains no model of its own. It is code that connects to models made by others and organises what happens around them: documents, tools, memory and the order of steps.
What is the difference between LangChain and LangGraph?
LangChain offers quick, ready-made building blocks and a simple way to create an agent. LangGraph is the lower-level library for designing every step yourself, as a graph. Both come from the same company, and LangChain's agents run on LangGraph.
Is LangChain free?
The libraries are free and open source under the MIT licence. The company earns money from LangSmith, a separate platform for monitoring, testing and deploying agents.
Do I need LangChain to build an AI app?
No. Many developers call a model's API directly. A framework like LangChain saves time when an app has many steps, documents or tools, at the cost of learning its way of doing things.
Was this clear?
1 is confusing, 5 is crystal clear. One tap, no account needed.
Spotted something wrong, or know a better example? Suggest a change