Producing Outputs

When the graph hits END, graph.invoke() returns the Final State Snapshot.

Identifying the Result

Usually, the state contains a lot of noise (intermediate tool calls, scratchpad thoughts).

result = graph.invoke(inputs)

# The result is the Full State Dict
# {'messages': [Human, AI, Tool, AI], 'scratchpad': 'thought...'}

# Extract what you show the user
final_msg = result["messages"][-1]
print(f"Agent Replied: {final_msg.content}")

Streaming Outputs

For chat apps, use streaming to show progress:

for event in graph.stream(inputs):
    # event contains the output of the node that just finished
    print(event)