What this does

The state schema defines what data flows through a LangGraph workflow. Each node reads from and writes to this shared state, with reducers controlling how updates are applied.

Steps

Define a basic state schema with TypedDict. Every key is a state field.

from typing import TypedDict

class SearchState(TypedDict):
    query: str
    results: list[str]
    error: str | None

Add reducer annotations for append semantics. Use Annotated to specify how updates merge.

from typing import Annotated, TypedDict
from operator import add

class ChatState(TypedDict):
    messages: Annotated[list, add]  # appends new messages
    context: str
    step: int

Use add_messages reducer for LangChain message objects. This handles message deduplication by ID.

from typing import TypedDict, Annotated, Sequence
from langgraph.graph.message import add_messages
from langchain_core.messages import BaseMessage

class AgentState(TypedDict):
    messages: Annotated[Sequence[BaseMessage], add_messages]
    next_agent: str

Define a default factory for complex types. Prevents key errors on first access.

from typing import TypedDict
from dataclasses import dataclass, field

@dataclass
class WorkflowContext:
    documents: list[str] = field(default_factory=list)
    score: float = 0.0

class AdvancedState(TypedDict):
    query: str
    context: WorkflowContext

Inspect state at runtime. Use NodeInterrupt or logging to see state shape.

def inspect_node(state: AgentState) -> dict:
    print(f"State keys: {list(state.keys())}")
    print(f"Messages count: {len(state['messages'])}")
    return {}

Verification

python -c "
from typing import TypedDict, Annotated
from operator import add
class S(TypedDict):
    items: Annotated[list, add]
s: S = {'items': ['a']}
s['items'] = s['items'] + ['b']
print(s['items'])
# Expected: ['a', 'b']
"

Common failures

Missing Annotated reducer. Without a reducer, updating a field replaces it instead of merging. This causes data loss in cyclic graphs.
Mutability side effects. If a node modifies a list in place, the reducer may not trigger. Always return new values.
KeyError on first access. Nodes try to read a key that hasn't been initialized. Provide defaults or use state.get("key", default).
Version mismatch - The installed package or runtime differs from the command shown; check the version first and rerun the smallest verification command.
Local environment drift - Another service, virtual environment, model, or path is being used; print the active binary path and configuration before changing the guide steps.

Related guides

How to Set Up Basic LangGraph Agent from Scratch
How to Implement Memory in LangGraph State