State schema

A state schema names the data that moves through the graph. Every node reads the current state and normally returns only the keys it changed.¹

Tiny example

from typing_extensions import TypedDict

class RefundState(TypedDict):
    order_id: str
    eligible: bool
    reason: str

def check_policy(state: RefundState) -> dict:
    return {"eligible": True, "reason": "Returned within 30 days"}

check_policy does not need to return order_id; LangGraph applies its partial update to the existing state.

Schema choice	Use it when
`TypedDict`	You want a light, typed dictionary
`dataclass`	You need defaults and a Python object
Pydantic model	You need recursive input validation and accept extra overhead

Failure note

TypedDict helps type checkers; it is not runtime validation. Keep secrets and large service clients out of state. State may be serialized into checkpoints.

LangChain, Graph API overview. ↩