A live meter for Claude Code that shows how full your context window is, so you /compact before your agent forgets your code.
One session, top to bottom. You see the drop coming before the answers go bad.
Early on, your agent one-shots the function. Three hundred messages later it re-introduces the bug you killed an hour ago and invents an API that does not exist. You cannot feel it coming. The meter can.
Empty window = sharp. Full window = mush. The bar names exactly where you are.
The real context-window percentage, straight from Claude Code, on your machine. No guessing, no network call.
Every failure path falls back to safe output and exits clean. 64 tests prove it.
MIT, no account, no telemetry, no config. Install it and it works immediately.
It reads a number Claude Code already computed. It never calls the model. It will never cost you a cent or a single token of context.
Paste three commands into Claude Code. The meter shows up.
The product is free and that is the honest answer. The other two exist because we could not help ourselves.