I'm implementing `open-deep-research` (<langchain ...
# koog-agentic-framework
r
I'm implementing
open-deep-research
(langchain code), and there is a "node" like we would say, `supervisor`: it holds top-level state of current research, topics, then delegates the specific topic for research to a standalone `researcher`(in parallel). How do you think guys, should it be implemented with
koog
like different agents or like nodes? I think this is better to hold in one agent, however not sure how to
forwardTo
in parallel. (supervisor implementation in python)
k
If you want to run "researchers" on the same machine - the subgraph per researcher might be the right option to start with. Check out the Trip planning agent example. It implements the similar pattern
v
r
[not Koog related] Omagad, couldn't find system_prompt for
supervisor
node in langchain implementation for a while. They are setting it in previous node T_T. And this is official repo with 9K+ stars ... Thanks God I can write AI Agents with `Koog`+`Kotlin` and not Python
kodee happy 1
(I understand, that we can also write statefull noodles, but its like Python best-practises)