Observed arrival · 2026-09-22
Tokenfatty: A Local Meter for AI Coding Costs
A macOS menu-bar meter that reads Claude Code’s local transcripts and estimates what the work would cost at public API rates.
- For
- Claude Code users tracking subscription usage
- Worth noticing
- The sample report separates cache reads, cache writes, uncached input, and output, rather than treating tokens as one undifferentiated total.
Field notes
The tool models subscription usage against public API list rates rather than presenting a bill. Its displayed accounting distinguishes cache writes from cache reads and output, including separate rates for each bucket. The homepage also describes a command-line report covering plan windows, daily and hourly timelines, project totals, and individual expensive responses. A local-first workflow is central to the presentation: the report is described as self-contained and usable with the network offline.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue