As we established in a previous conversation in th...
# koog-agentic-frameworka
ait
06/23/2025, 11:28 AMAs we established in a previous conversation in the server environment we need pool of agents.
But in real life agent should be also aware about current user/chat context. Any advice except se passing JSON as agent input?
An the next question is how to pass sort of "agent state", for example when agent was waiting (logically) user input/answer
d
Didier Villevalois
06/23/2025, 12:05 PMYou will have to encapsulate your session state into an object. For example, the current prompt, and maybe other state like for instance the selected tools (if the user can choose those), etc.
Then use this session state to correctly initialize your agent at each request and retrieve the corresponding data after the agent ran to update your session state.
C.f. https://github.com/JetBrains/koog/issues/218
m
Mark Tkachenko
06/23/2025, 2:43 PMWell, it really depends on your scenario. Do you run different agents on each user message or you’re trying to make chat-like conversation with single agent? Are you trying to reuse running agents for different users?
Be advised that in the last case you might collide with different users history.
And yeah, @Didier Villevalois is right - it really looks like scenario for chat-like agent