Token spend per actor, three separate time measures, the parallel-agent flow timeline, and every sub-agent task as a searchable row — from the transcript Claude Code already wrote on your machine.
Privacy First Runs in your browser — the file never leaves it
EN · 中文 · TR — the interface and every report ship in three languages
other languages welcome — contributions are open
Input, cache-creation, cache-read and output per actor — the main thread and each sub-agent type — with the metric switchable.
Wall-clock, active wall-clock (idle removed) and agent work-hours (parallel agents summed). Mixing these up is how session numbers get misread.
Lanes of what ran concurrently, concurrent-agent count, tokens over time, and the run split into activity blocks on real idle gaps — 20 minutes with no API call and no running sub-agent.
English · Türkçe · 简体中文, light and dark. The downloaded report is one self-contained HTML file that works with no network at all.