LangChain
The most widely adopted open-source framework for building LLM applications — chains, agents, retrieval, and integrations across hundreds of providers — and the ecosystem around LangGraph and LangSmith.
Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.
What it is
LangChain (2022, Python and TypeScript) packages the building blocks of LLM applications: prompt templates, retrieval, tool integrations, memory, and model adapters for effectively every provider. Around it sit LangGraph, for stateful agent orchestration, and LangSmith, for tracing and evaluation.
Why it matters
LangChain is the de facto entry point to LLM engineering and the largest integration catalog in the ecosystem. It is also regularly criticized for abstraction depth: teams often prototype on it and later slim down to direct SDKs or lighter frameworks. Knowing when it earns its place is part of the engineering decision, not a default.
How we use it
Our default agent framework is Pydantic AI. When an engagement's existing stack is LangChain or LangGraph, we extend it rather than rewrite it — the runtime contract we add (versioning, evals, traces, promotion gates) is the same on either substrate.
Related resources
LangChain's library for building stateful agent workflows as graphs — nodes, edges, checkpoints, and human-in-the-loop interrupts — one of the common substrates for production agent orchestration.
The agent framework from the Pydantic team — typed agents, structured outputs, tool calling, retries, and evals built around Pydantic validation, with first-class multi-provider and fallback support.
The scaffolding around an AI agent — prompt construction, tool dispatch, retry logic, trace emission, state management — that turns a model into a workflow participant.
The pattern where an AI agent retrieves relevant context from your data before generating an answer — instead of relying only on what the model learned during training.