LangChain
LangChain is “a framework for building agents and LLM-powered applications”. It gives you one standard interface to models, embeddings, vector stores and more, so you can swap providers without rewriting your app. It’s MIT-licensed. The Python package is at version 1.4.7 (released 21 October 2026) and needs Python 3.10 or later; LangChain.js covers JavaScript and TypeScript.
The core idea: create_agent
The docs describe create_agent as “a minimal, highly configurable harness”: everything around the model loop, meaning the prompt, the tools and any middleware that shapes behavior. You give it a model, tools and a system prompt, and add only the middleware your app needs.
Where it sits in the stack
- LangGraph is the lower-level orchestration framework. LangChain’s agents are built on top of it, which gives them durable execution, human-in-the-loop support and persistence. Use LangGraph directly for workflows that mix fixed steps with agent steps.
- Deep Agents is a batteries-included harness built on LangChain agents, adding planning, subagents, context compression and a virtual file system.
- LangSmith is the paid platform for tracing, evals and deployment. Its Developer plan is free for one seat with 5,000 traces a month, and Plus costs $39 per seat per month (pricing, checked 1 Nov 2026).
Who it’s for
- Developers building agents or RAG apps in code who want to switch model providers freely.
- Teams heading for production, who can move to LangGraph and LangSmith without changing ecosystems.
Compare
- LangChain vs LangGraph: which layer you need.
- Dify vs LangChain: visual platform or code framework.
- CrewAI vs LangGraph vs AutoGen: agent frameworks side by side.