tmux and Claude Code

How an old technology is helping me navigate the AI Era

What is tmux

If you’re not familiar with tmux, it is short for “terminal multiplexer.” I have previously written a post on it that you can refer to.

Claude Code

Claude Code being a TUI (terminal user interface) was one of the things that got me excited to use it. As someone who spends most of their time in the terminal - at least when working on code - tools like Cursor were not as appealing to me. The terminal actually seemed like a natural fit in my mind - everything is text-based, which is perfect for a language model. While some of my colleagues were less comfortable and wanted a GUI application, I was very happy to keep things as text.

Setup

The main thing for me is that it is a natural fit in my pre-existing flow. I typically have one Window per repo or project, with a pane on top for editing (Neovim), and multiple other panes down below - usually one for running the app (e.g. npm run dev), and one for dealing with Git (shoutout to lazygit). This meant that I could easily slot in Claude to one of those panes, or add another one.

Result

This is a pretty simple setup, and what I imagine many others are doing too, but what it means in practice is that it is trivial for me to have a Claude Code session per repo - or even per Git worktree if needed. Despite what I see many people talk about online, I actually try to focus in on one session at a time, so that I can keep my focus and understanding in one place; this is likely because I am less trusting and still reviewing code myself, at least before committing. But, even then I still see a benefit: I can have separate contexts for each project as I have to jump around for various tasks throughout the day, which means fewer compactions or context resets.

In any case - it is nice to see gains from using relatively old tools (tmux is actually not that old, coming out in 2007) that I’m already comfortable with, even in the age of AI.