stream() and astream()

stream() and astream() expose progress before the final result is ready. Pick a stream mode based on what the client needs.¹

Tiny example

for part in graph.stream(
    {"question": "Check order A-104"},
    stream_mode="updates",
    version="v2",
):
    if part["type"] == "updates":
        print(part["data"])

Mode	What it exposes
`updates`	Node name and its state update
`values`	Full state after a step
`messages`	LLM message/token chunks with metadata
`custom`	Application-defined progress events

Use async for part in graph.astream(...) in an async application.

Failure note

Streaming is an observation interface, not a checkpoint. A client seeing an update does not by itself mean the update is durably stored. Configure a checkpointer and durability mode separately.

LangChain, Streaming. ↩