Continuing Agent Conversation in different Scenario with new Agent Framework

Questions
1

:bullseye: What is your goal?

I would like to reuse an agent I build with the new Agent interface in another scenario. So far, with the old make Agents, I could create an agent in the Agents tab, choose it in my two different scenarios, and by using the same Thread ID, I could continue the same conversation.

:thinking: What is the problem & what have you tried?

I do not see this possibility with the new Agent design. I could copy and paste the node, but then it would create a new agent, even when using the same Thread ID.

:clipboard: Error messages or input/output bundles

I can not reference an Agent I created in one scenario, in the other.

3

Hey Marja,

put the agent in its own scenario started by the Start Scenario module and ending with the Return output module (if you need the agent’s output back for further processing). And every time you need to call the agent from within another scenario, use the Call a Scenario module.

Inside both Call a Scenario and Start Scenario modules you can define the data that will be sent from the master scenario to the agent scenario.

4 Likes
4

Hello Stoyan, thank you for this idea, it works! It is more of a workaround though, I hope that make will provide a solution that is as simple as the original one.

5

This is technically the same as the original solution. Except now you can see what the agent is doing and debug it.

With the original layout it essentially worked the same way as I described, except the agent actions were hidden and you couldn’t track them. You’d use call an agent module instead of call a scenario and you’d configure the agent from a separate menu instead of the scenario canvas, but technically it wasn’t that different.

1 Like
6

You are right, technically, it works the same now. For me though, it was more convenient to configure the agent in a separate tab and then reference in any scenario I want with the simple drop-down. Now, you have the overhead of creating an extra scenario for an agent, if you want to be able to prompt it multiple times in the same conversation, but not the same run.