LangGraph

LangGraph is a library from LangChain for building stateful, multi-actor applications with LLMs. It extends LangChain's capabilities by enabling the creation of complex agentic workflows with cycles, branching, and persistent state management.

Unlike simple chains that flow linearly, LangGraph allows building applications where agents can loop, branch based on conditions, and maintain state across interactions. Key concepts include nodes (functions or agents), edges (connections with optional conditions), and state (shared data that persists across steps).

LangGraph is particularly useful for: building multi-step agents that need to iterate, creating workflows with human-in-the-loop checkpoints, implementing complex reasoning patterns like tree-of-thought, and orchestrating multiple specialized agents. It integrates with LangSmith for observability and debugging of complex agent workflows.