Performance

The levers to adjust CPU consumption and API call frequency to your setup.

Refresh frequency

TokenEater polls the Anthropic API every 5 minutes by default. It's configurable in Settings -> Performance, with a minimum of 3 minutes.

The reset countdown in the menu bar updates every minute without hitting the API. So you can have a smooth timer even with slow polling.

CPU usage

TokenEater includes a few real-time features (Agent Watchers, animated gradient background) that can slightly raise CPU usage. If that's a concern, several levers:

Disable Agent Watchers if you don't need session tracking
Disable the animated gradient background
Increase the API polling interval (5 min -> 10 min or more)

Cache strategy

The per-file mtime cache in History avoids re-parsing already-known Claude Code sessions. JSON parse is streaming, with a parallel TaskGroup capped at a reasonable concurrency.

First open of History builds the cache, subsequent opens are instant even on multi-month projects.

On MacBook Airs or more modest machines, disabling the animated gradient background often makes the most visible difference. Agent Watchers themselves consume very little, except during intense multi-session activity.