Send and parallel fan-out

Send creates a dynamic call to a node with its own input. Return several Send objects to fan work out in parallel, then merge results through a reducer.¹

Tiny example

import operator
from typing import Annotated
from langgraph.types import Send

class ResearchState(TypedDict):
    urls: list[str]
    summaries: Annotated[list[str], operator.add]

def fan_out(state: ResearchState):
    return [Send("summarize", {"url": url}) for url in state["urls"]]

def summarize(state: dict) -> dict:
    return {"summaries": [fetch_summary(state["url"])]}

1. Router sees three URLs.
2. It returns three Send("summarize", ...) values.
3. Three node calls run concurrently.
4. The reducer combines their summaries.

Failure note

Cap fan-out and invocation concurrency. Parallel calls can overload an API, and results should not rely on completion order. Side-effecting workers need idempotency keys.

Related

• Reducers
• Edges
• Async waiting and timeouts

Footnotes

1. LangChain, Use the graph API — Map-reduce and the Send API. ↩