# QMix Grouping Agents in ExternalEnv Configuration

**URL:** <https://discuss.ray.io/t/qmix-grouping-agents-in-externalenv-configuration/9707>\
**Category:** Configure Algorithm, Training, Evaluation, Scaling\
**Created:** [March 9, 2023, 5:48pm UTC](https://discuss.ray.io/t/qmix-grouping-agents-in-externalenv-configuration/9707 "2023-03-09T17:48:23Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![hermmanhender](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ray.io/hermmanhender/32/5691_2.png) [@hermmanhender](https://discuss.ray.io/u/hermmanhender)\
**Post date:** [March 9, 2023, 5:48pm UTC](https://discuss.ray.io/t/qmix-grouping-agents-in-externalenv-configuration/9707/1 "2023-03-09T17:48:23Z")

</div>

Hi!  
I’m trying to configurate a QMix algorithm in RLlib 2.3.0.  
Previously I configured a DQN algorithm and worked good, but now I need to my agents cooperate between them. I read that QMix is a good alternative for a centralized training and a descentralized `get_action()`. Can you confirm that? There are a better alternative?

But my question here is the follow: **How can I configure a grouping of agents in a Server/Client (or ExternalEnv) configuration?**  
In [another post](https://discuss.ray.io/t/playing-the-qmix-two-step-game-on-ray/4200/4) it was tell that you need to define your environment, but in the configuration that I need I have not an environment, as the case in [this post](https://discuss.ray.io/t/externalmultiagentenv-and-qmix-for-remote-inference-over-http-with-multiple-clients/3695), where was suggest that the grouping must be in the PolicyClient side. I don’t really understad that final suggestion. There are an example of QMix algorithm with None environment?  
Thanks!