I’m using the Open Deep Research LangGraph agent. I want to capture sources and activities that are produced inside subgraphs and persist them - ideally without using an interrupt.
My setup
-
Graph compiled with a checkpointer (
MemorySaver) +thread_id. -
Running on LangGraph Platform.
-
A FastAPI service is polling/querying the LangGraph Platform API to save sources and activities (of the research) in the database and show them in the UI.
What I’m unsure about
The research can be long (10-20 minutes)
With how this agent is built is only has a few nodes
clarify_with_user → write_reserach_brief → research_supervisor → final_report_generaion.
So currently, I can only get state updates in the UI after each of those finishes.
But the issue is that all the magic happens inside the research_supervisor subgraph.
How can I get its state during a run?