LangChain

Teams writing LLM features in Python reach for LangChain when they want one programming model across many model vendors. Version 1 centers on create_agent, a compact harness pairing a chat model with tools, a system prompt and stackable middleware hooks for retry logic, guardrails, model routing or tool policies. Those agents execute on the LangGraph runtime, so checkpointed state and human approval steps come without extra wiring. The monorepo holds langchain-core with the shared abstractions, text splitters, a standard test suite, and partner packages for OpenAI, Anthropic, Ollama, Groq, Mistral, Qdrant, Chroma and others, while more providers, such as Google and Azure, ship separately. Install it with pip or uv on Python 3.10 or newer; nothing runs locally unless you point it at a local model server. The code is MIT licensed and free for commercial products; LangSmith, the vendor's tracing and deployment service, is optional and commercial, with a free single-seat tier for tracing. A good fit for developers who value swappable providers and a broad connector catalog, less so for anyone chasing a minimal dependency tree.