I'm open sourcing my tmux configuration to quicly switch between agents, monitor progress and overall be able to manage 10+ agents easily on a remote server.
I feel it is a good sign when friends ask you to share your custom setups when you're showing them something, so I thought it was time to share this publicly too.
Getting started
https://github.com/Ymirke/tmux-agent-switcherBackground
The last 6 months has been a crazy time to work on software, most people I know rarely write any code by hand. This for a while has created what I have been calling "brainrot coding", where you micromanage a ton of agents to achieve what you want.
The setup is typically similar to what conductor and others now offer, where each agent session has its own git worktree or its own branch. This means you typically setup two scripts setup.sh and run.sh which setup your environment and run it. The reason why you often need these additional scripts is because they mange it so that none of your enviornments have colliding ports with eachother. It is now pretty common to jump around on different versions of the software you're building on your own machine, while AI is building for you.
This tool doesn't do this, but it helps you switch between sessions quicker. The way I use it is to have one tmux session per feature branch, and then spin up agents as needed to make the changes I want.
I'm also happy that I'm moving more and more away from having a ton of agents running, I attribute this to the models becoming much better at long running tasks, and part of this is being much better at scoping out changes before they start. In any way, it feels good to have this at a point where I can open source this ( and people can use it) almost as a tribute to this somewhat crazy six months.
This is something I think a lot of AI products should think about, why would any user want to manage may agent sessions. I just want to have the main agent manage my agents.