LangGraph

LangGraph is a low-level orchestration library for building agents as explicit state graphs: you define nodes (steps), edges (transitions), and a shared state object, and the agent's control flow becomes something you can see, test, and resume. It trades the one-line convenience of higher-level frameworks for control — which is exactly what production agents tend to need once they outgrow a demo.

It is aimed at engineers building durable, multi-step or multi-agent systems where you care about persistence, branching logic, and being able to pause for human input. Despite the name, LangGraph does not require the rest of LangChain.

Highlights

• Graph-based control flow — model loops, branches, and multi-agent handoffs as explicit nodes and edges instead of opaque prompt chains.
• Persistence / checkpointing — save and restore agent state, so runs are resumable and crash-safe.
• Human-in-the-loop — interrupt the graph for approval or input, then resume from the exact point.
• Streaming — stream tokens and intermediate steps for responsive UIs and debugging.
• Deployable — pairs with LangGraph Platform for hosted deployment, plus LangSmith for tracing.

In an AI-assisted workflow

from langgraph.graph import StateGraph, START, END
 
g = StateGraph(State)
g.add_node("retrieve", retrieve)
g.add_node("generate", generate)
g.add_edge(START, "retrieve"); g.add_edge("retrieve", "generate"); g.add_edge("generate", END)
app = g.compile(checkpointer=checkpointer)  # resumable, interruptible

Good to know

LangGraph is open source (MIT) and free to self-host; the optional LangGraph Platform (hosted deployment) and LangSmith (observability) are commercial. It's lower-level than role-based frameworks, so expect to write more wiring in exchange for more control.