mirror of
https://github.com/langchain-ai/langgraph.git
synced 2026-08-31 04:09:49 +02:00
Replace the passthrough (chunk, metadata) tuple projection with one that yields a ChatModelStream per LLM call, routed by run_id. Handle both v2 protocol-event payloads (message-start/chunk/message-finish) and whole AIMessage payloads from on_chain_end (replayed via message_to_events). Wire _bind_pump from GraphRunStream so nested sync streams share the caller-driven pump.