LangGraph

LangGraph is an open-source library for orchestrating LLM agents as explicit, stateful graphs, giving developers fine-grained control over loops, branching, checkpointing, and human-in-the-loop steps. Maintained by LangChain, it is free to self-host under an MIT license and works with any model provider, while optional paid products (LangGraph Platform for deployment, LangSmith for observability) sit on top.

LangGraph models an agent as a graph of nodes (functions or LLM calls) connected by edges, with a shared, persisted state object flowing between them. This lower-level, more explicit approach was designed to fix a common complaint about earlier LangChain chains and agents: that they were hard to control, resume, or debug once loops and branching were involved. Because the graph is explicit, developers can add cycles, human-in-the-loop approval steps, checkpointing, and time-travel-style replay of a run. The core library is MIT-licensed and free to self-host, and it works with any LLM provider rather than locking you to a specific model. Around it, LangChain sells commercial layers: LangGraph Platform for managed or self-hosted deployment of agents as APIs, and LangSmith for observability, evaluation, and prompt management. These are separate products with their own pricing, so you can use the framework entirely for free and only pay if you adopt the hosting or tracing services. LangGraph has become one of the more widely adopted agent frameworks for teams that want production control rather than quick prototypes. It is a heavier, more code-first tool than higher-level agent builders, and it assumes comfort with graph and state concepts, which is the main trade-off for the control it provides.