LangChain

LangChain provides a common set of abstractions — chains, agents, retrievers, memory — for building applications on top of LLMs, so switching model providers or adding retrieval-augmented generation doesn’t mean rewriting your application logic each time. Its scale (700+ integrations) is both the appeal and the common complaint: broad provider support, but a larger surface area to learn than a narrower tool.

If you specifically want programmatic prompt optimization rather than a general application framework, see DSPy — a different approach to a related problem, not a direct substitute.