LangGraph

LangGraph is a free, open-source developer library for building AI agent workflows as an explicit graph of steps. Developers use it to design single-agent, multi-agent, or hierarchical workflows with lower-level control than LangChain's higher-level patterns. They can add human-in-the-loop checkpoints, where a person must approve a step before the agent continues, persist memory across sessions, and stream an agent's reasoning and actions to a user in real time. It comes from the same team behind LangChain, under an MIT license, and works with any model provider. LangSmith is a separate paid product from the same company that adds monitoring and a hosted deployment option called LangGraph Platform.

This fits an engineering team that has outgrown a simple agent pattern and needs explicit control over branching logic, retries, or approval steps, such as a loan review or a multi-department support workflow. Start by mapping one real process as a graph on paper before writing code, since the library's value comes from that explicit structure. The library itself has no license cost; expect to pay for underlying model API calls, and for a separate LangSmith or LangGraph Platform subscription if you want built-in monitoring or hosting, which the vendor prices rather than publishing. The tradeoff for this level of control is more setup work than a no-code agent builder, so plan for real developer time before launch.