I kept running into limits on my Claude Code sessions (Especially with leading edge models)... the main cause being session hygiene, not clearing or compacting my sessions often enough.

So I built this simple tool to show my current session status.

My two cents: the README should feel effortless to read and easy to skim. For something this straightforward, I’d ideally want to grasp the key points without having to scroll much. Once it starts feeling longer than it needs to be, I tend to lose interest and I think I'm not the only one.

I built something similar but as a tray/cli and not just for usage. You can switch the login of CLIs like Codex, Claude, Vercel (if you're managing multiple accounts) etc

Here's Switcheroo: https://github.com/FTCHD/switcheroo

Unaffiliated with this project but I enjoy Claude Code Usage on my Mac https://github.com/richhickson/claudecodeusage

Just use autocompact?

Insane slop title.

> context a session moves grows with roughly the square of its length

Readme doesn't mention that LLM providers do context caching — your costs aren't quadratic unless you frequently invalidate the cache.

What's funny is that, being a claude code user, I don't need to give someone else's python package a 24/7 back door into all my activity, I can just make my own.

I never have long sessions. I ask claude to do one thing, then I clear the context, validate that change, and iterate it. What are these long sessions?

Someone about to learn about caching.