LangChain is the framework most developers think of first when building an LLM-powered application, largely because it got there early and built integrations for practically every model provider, vector store, and tool a developer might want to wire together. It wraps the repetitive parts of building with LLMs, prompt templates, memory, tool calling, and chaining multiple calls together, into reusable components instead of one-off glue code.
With its 1.0 release in October 2025, LangChain standardized its agent-building API around a create_agent function and a middleware system, and every agent built this way now runs on LangGraph's execution runtime underneath. That consolidation reflects a broader shift in the ecosystem: LangChain used to be mostly about chains and prompt orchestration, and is now explicitly an agent framework with LangGraph as its execution engine.
It's free and open source, with over 100 model provider integrations, both its biggest strength and a recurring complaint: breadth sometimes comes at the cost of a learning curve, since there's rarely just one obvious way to do something in LangChain.