mirror of
https://github.com/langchain-ai/langgraph.git
synced 2026-08-25 00:52:25 +02:00
Replace the daemon thread pump with a pull-based model where the caller's iteration on any projection drives the graph forward. EventLog uses a _request_more callback instead of threading.Condition. Matches v1's model where the caller's for loop is the pump. Async path is unchanged (background task on the event loop).