Glossary
What is LangChain?
In short
LangChain is an open-source library for building AI agents and applications on language models, distributed under the MIT license.
Models and products · LangChain
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine a box of toy railway parts: rails, switches, little cars, traffic lights. On their own they just lie there, but connect them and the train goes where you need it to. There is a set of parts for programmers that they use to build smart helpers: one piece listens to the question, another looks for the answer in books, a third writes the answer. You have to put them together carefully so the train does not come off the rails.
In plain words
It is a set of ready blocks for programmers who make applications with AI. Instead of writing the connection to a model, the search through documents and the calls to outside services from scratch each time, a developer takes ready parts and connects them. One program can work with models from different providers, so you can change the model almost without reworking the code. Example: a support chatbot that finds an answer in the knowledge base and, when needed, checks the order status in the accounting system.
For business
The library itself is free and open under the MIT license: it is used in commercial products and can run entirely on your own servers, including with a local model through Ollama. The costs come not from it but from the chosen model, the infrastructure and the developers’ work. For a business, another benefit matters: changing the model provider does not require rewriting the agent, so dependence on one company is lower. Cloud services around the library, such as LangSmith for observation and evaluation, are separate products with their own terms, and a lawyer should check them. Before an agent enters work processes, set up access rights to systems and an action log: these questions are decided when you design an AI division.
For a specialist
The documentation defines an agent by the formula “model plus harness”: a system prompt, tools and middleware. The main entry point, create_agent, builds such a harness, and the agents themselves are built on LangGraph, a low-level orchestration layer with durable execution, state persistence and human in the loop. A standard interface covers chat models and embeddings from different providers; examples in the documentation include OpenAI, Anthropic, Google, Ollama and Hugging Face. The loop in which a model alternates reasoning and tool calls is described in the ReAct paper (Yao et al., 2022). The core library is written in Python, and a separate repository, LangChain.js, covers JavaScript and TypeScript. Next to it, LangSmith is developed for evaluation and observability, and Deep Agents for planning, subagents and work with files.
Example
A developer at an online store uses the library to build an agent that looks for answers in the knowledge base and checks the order status through an API. When the company decides to change the model, the developer changes the connection setting, and the agent’s logic stays the same.
Related terms
Questions and answers
Sources
Where companies use it
Not sure whether your company needs this?
AI solutions map