LangChain

LangChain is the earliest mature LLM application framework, defining abstractions (chains, agents, tools, retrievers) that are still widely reused today. It standardizes how LLMs compose with external tools, memory, and vector stores, and is a common starting point for learning LLM application development.

Core capabilities cover prompt templates, memory modules, document loading and splitting, vector retrieval, agent tool calling, LangGraph state-machine orchestration, and the LangSmith observability platform. The ecosystem integrates over a hundred model and data-source providers, so almost anything you can think of has an integration.

The highlight is a huge ecosystem with rich tutorials, and almost any provider has a ready-made integration. The downside follows from the same root: abstraction layers are thick, versions change frequently, official docs often lag behind code, and the 0.x to 1.x migration cost is non-trivial. Debugging deep chains can be painful.

Use cases: rapid RAG prototyping, applications needing multi-provider switching, teaching and course experiments. Evaluate the controllability cost of abstractions before production deployment.