Streaming Output AI Agent node process ，Instead of synchronously outputting after the node ends

Feature Requests
,
1

The idea is:

Streaming Output AI Agent node process ，Instead of synchronously outputting after the node ends.Capable of real-time output of agent processing procedures

My use case:

I think it would be beneficial to add this because:

Should observability control continue or provide guidance

Any resources to support this?

Are you willing to work on this?

2

The last entry in the dropdown

image
image757×753 41.3 KB

and the following doc make me believe that it’s already a thing?..

3

At present, the agent flow only outputs the flow of “finish.rason” and “stop”, without any intermediate processing or thinking processes

4

I don’t know exactly what are you expecting, but I get definitely the “intermediateSteps” of what tools triggered with all the args and steps…

Can you elaborate with an example on what you are expecting?

5

When the agent node starts, it begins to stream the intermediate process results, such as: the first step is to call the model, the model responds to the tool call, the second step is to call the tool, the tool responds to the result, the third step is to call the model, the model responds, the fourth step is to call the tool, the tool responds to the result, and the fifth step is to call the model, the model responds to the final result;
It is a real-time streaming output process and final result, not just a streaming output of the final result