free · open source · runs on your machine

Does your AI agent get worse as its context fills?

contextrot reads your agent's logs and tells you.

the story ↓

01Everything your agent sees goes in here.

01Files. Messages. Tools.
It fills.

01Past 60%, they say,
it starts to rot.

02

So we counted
every slip.

scroll →

01edit missed
02same call twice
03re-read a file
04tool error
05"sorry, let me fix that"

At every fill level.

03We checked
31,991 real steps.

No rot.

Filling the window wasn't the problem.

04

So what was?

← scroll

1.6×

late nights

5.1% 0–6h3.2% 6–12h

4.5×

which agent CLI*

14.8% opencode3.3% claude code

8.5×

which model*

22% worst2.6% best

1.0×

mistakes piling up

no effect

1.0×

long solo runs

no effect

Yours will be
different.

* different tasks · association, not causation

Find out in seconds.

demo data

click to copy · Python 3.9+ · MIT · zero network calls

Claude Code · Codex · Antigravity · Gemini CLI · Qwen Code · OpenCode · Cline · Roo · Kilo