is it possible to update state in condition function?

[anmol-aidora]

Can I update the state in the condition/path function of graph.add_conditional_edges?

def condition(self, state):
  state["variable"] = "UPDATED"
  return random.choice(["OPTION_A", "OPTION_B", "REPEAT"])

graph.add_conditional_edges(
  "node_a",
  self.condition,
  {
    "OPTION_A": "node_b",
    "OPTION_B": "node_c",
    "REPEAT": "node_a",
  },
  )

I have tried but its not working. If it doesn't work, how do I make it work? One option for me is to write a new node to process the state but it will increase redundancy.

[edwinow]

Hi Anmol, I don't think you can unfortunately, and it would mix up the concept of nodes and edges.

Would it be possible instead to

• either update the state either the preceeding node (if you don't need the outcome of the random choice)
• or in the destination node (in case the random choice impacts how you update the state)?

You won't have to write a new node as the existing nodes can update more than 1 piece of state (can add more than 1 key value pair in the return dict of the node).

def node_a(state):
    response = llm.invoke(state['messages']) 
    return {"messages": [response], "variable":"UPDATED"} #update state['variable'] in existing node

[anmol-aidora]

Thanks @edwinow ! I have done something similar. I was problem with capturing human feedback. I was trying to take human feedback in previous node (A), add that node in interrupt_after and process the human feedback in next node (B). But it wasn't working.

I then decided to add a no-op function just to take human feedback (H), and added it in interrupt_before, and now its working.

So I went from A->B to A->H->B.

But I still wonder, why it wouldn't work in A-B configuration. I was already adding interrupt_after in A. Adding B to interrupt_before also didn't work.