LangChain

LangChain is the open-source Python (and JavaScript) framework that pioneered composable LLM applications — chains, prompts, tools, memory, and RAG — and became the default starting point for a generation of AI apps. Version 1.x repositioned it as an agent engineering platform with a streamlined agent API, deep agents (planning, research, sub-agents), conversational memory, structured outputs, and production deployment paths.

It is model-agnostic (OpenAI, Google Gemini, Anthropic, OpenRouter, Ollama, Bedrock and more via provider integrations) and pairs with LangSmith, its observability companion. Tracing to LangSmith is strictly opt-in (LANGSMITH_TRACING=true plus an API key) — with tracing disabled, no user data is persisted externally.

1. Install the package: pip install langchain (add your model provider's integration package, e.g. for OpenAI / Gemini / Anthropic, as shown in the quickstart's provider tabs)
2. Get an API key from any supported model provider (OpenAI, Google Gemini, Anthropic, OpenRouter, Ollama, AWS Bedrock, Hugging Face, and others)
3. Export the key in your shell or put it in a .env file — e.g. export OPENAI_API_KEY=<your-key> — and load the file with python-dotenv (load_dotenv())
4. Build the quickstart agent from the docs: start with the basic tool-calling agent, then the real-world research agent to explore system prompts, custom tools, memory, and deep agents
5. (Optional) To inspect agent runs, sign up for LangSmith and set its environment variables to start logging traces — tracing stays off unless you explicitly enable it