dsh-token-stats

Every conversation in DeepSeek Harness consumes input, output, and cache-read/write tokens, yet those numbers are scattered across individual session logs. Answering simple questions like how much you have used in total, which models dominate, and which days were heaviest used to mean digging through raw records. dsh-token-stats aggregates token usage from all local sessions into a single live dashboard, making your consumption instantly legible.\n\nFrom the settings page you get a cumulative overview (input / output / cache breakdown), peak single-request and peak daily figures, streak tracking, and call-count statistics. Daily and weekly trend curves, a 24-hour distribution histogram, and a GitHub-style yearly activity heatmap reveal your usage rhythms. A donut chart with per-model details shows share, call count, and token split for every model. A persistent Today widget at the bottom of the sidebar gives hour-by-hour model stacking with a day-over-day comparison, and it can be detached into a draggable floating window so the sidebar stays uncluttered.\n\nAll figures are computed in real time by scanning local session logs, with no separate persistence layer and no data leaving your machine. Duplicate counting from forked, resumed, or sub-agent sessions is handled transparently. Animations respect the system reduce-motion preference and play only on first open; incremental updates land silently. Ideal for developers who call DeepSeek heavily, teams that need periodic token-cost reporting, and the simply curious who want to see how many tokens a single day burns.