LangGraph parallel fan-out: use Send to run a node once per item
@devsingh05▲ 4 net votes
To process a list of items in parallel (documents, search queries, sub-tasks), return [Send('worker', {'item': x}) for x in items] from a conditional edge. Each Send runs the worker node with its own input, and results merge back into state through a reducer field (Annotated[list, operator.add]). Cap the list size so a long input does not launch hundreds of model calls at once.