LangGraph
By LangChain
LangGraph is a framework from the LangChain team for building stateful, multi-step AI agent and LLM workflows modeled as graphs of nodes and edges, enabling more controllable and debuggable agent behavior.
Definition
LangGraph is a framework from the LangChain team for building stateful, multi-step AI agent and LLM workflows modeled as graphs of nodes and edges, enabling more controllable and debuggable agent behavior.
Overview
LangGraph extends the LangChain ecosystem by letting developers define an application's logic as a graph, where each node represents a step (such as calling an LLM, invoking a tool, or making a decision) and edges define how control flows between steps, including loops, branches, and conditional paths. This graph-based structure gives developers more explicit control over agent behavior compared to more open-ended, free-form agent loops. Because state persists across the graph, LangGraph is well suited for building agents that need memory, human-in-the-loop checkpoints, or complex branching logic — for example, an agent that retries a failed step, asks a human for approval, or routes a task to different sub-agents. It is often used alongside retrieval tools like LlamaIndex and competes with other multi-agent orchestration approaches such as CrewAI and Semantic Kernel.
Key Features
- Graph-based modeling of multi-step agent workflows
- Explicit control over branching, looping, and conditional logic
- Persistent state across workflow steps
- Support for human-in-the-loop checkpoints
- Built on top of the LangChain ecosystem