Nodes

A node is a Python function that owns one unit of work. It reads state and returns an update. The function may be synchronous or asynchronous.¹

Tiny example

def classify_ticket(state: TicketState) -> dict:
    urgent = "production" in state["message"].lower()
    return {"priority": "high" if urgent else "normal"}

builder.add_node("classify_ticket", classify_ticket)

Good node boundaries match failure boundaries:

1. Read one input.
2. Perform one understandable operation.
3. Return a small update.

Failure note

If the last operation in a large node fails, the node attempt may run again from its beginning. Separate external services into separate nodes, and make writes idempotent.²

Related

• Edges
• RetryPolicy
• Async waiting and timeouts

Footnotes

1. LangChain, Graph API overview. ↩

2. LangChain, Thinking in LangGraph. ↩