LangGraph
Stateful, graph-based orchestration for long-running agents
- Best for
- Long-running agent workflows with branching and persistence
- Pricing
- Open-source SDK; LangGraph Cloud has usage-based pricing
- Stack layer
- AI Stack
Live preview of langchain-ai.github.io/langgraph
Overview
LangGraph models agents as state graphs instead of linear chains. Nodes are work units, edges are transitions, and the framework persists state at every step. This is the difference between an agent that crashes mid-task and one that resumes exactly where it left off.
We pick LangGraph when the workflow has real branching — retries, conditional routing, human-in-the-loop approvals — that a linear chain would struggle to express clearly.
LangGraph Cloud adds managed persistence and streaming for production deployments.
Key Features
State Graphs
Model agents as nodes and edges with typed state
Checkpointing
Resume any workflow from any step
Human-in-the-Loop
Pause for approval mid-execution
Streaming
Stream tokens, state updates, and tool calls in real time
Cycles
Express loops that linear chains can't
LangSmith Tracing
Deep observability for every run
Why We Recommend LangGraph
State machines are the right abstraction for non-trivial agents. LangGraph gives you that without writing your own checkpointer.