# Agent guide: @bicrick's token usage This repo holds @bicrick's AI coding token usage: daily totals per tool and model, per machine, written by the tokentracker CLI. Do not edit files in `machines/` by hand. ## Answer questions from the static files (GitHub Pages, no server) https://bicrick.github.io/tokentracker-data/api/v1/stats.json totals, active days, busiest day, streaks, today, this month, top tool and model https://bicrick.github.io/tokentracker-data/api/v1/usage.json daily usage split by tool (also usage-week.json, usage-month.json) https://bicrick.github.io/tokentracker-data/api/v1/models.json per model: tokens, responses, share, first/last used https://bicrick.github.io/tokentracker-data/api/v1/tools.json per tool https://bicrick.github.io/tokentracker-data/api/v1/heatmap.json calendar grid (also heatmap-365d.json, heatmap-90d.json) https://bicrick.github.io/tokentracker-data/api/v1/rows.json every daily row https://bicrick.github.io/tokentracker-data/api/v1/index.json every file Time ranges are folders: https://bicrick.github.io/tokentracker-data/api/v1/{ytd,30d,90d,365d}/.json, e.g. https://bicrick.github.io/tokentracker-data/api/v1/30d/usage-week.json For anything else (from/to dates, tools=a,b, models=x,y, group=, split=, format=csv), the query layer computes from the same rows.json and stores nothing: https://tokentracker.patrickbrown.dev/api/v1/users/bicrick/usage?from=2026-11-01&group=week&split=tool (resources: usage, models, tools, heatmap, stats; profile at https://tokentracker.patrickbrown.dev/api/v1/users/bicrick) Semantics: tokens = input + output + reasoning + cacheRead + cacheWrite. responses = model calls. Dates are local days. Cursor has totals only (model "cursor/all"). Full reference: https://tokentracker.patrickbrown.dev/llms.txt ## Keep it working - A machine stopped syncing (an issue says so, or stats look stale): on that computer run `npx @bicrick/tokentracker@latest init --yes`. Safe to repeat; nothing is counted twice. - Add another computer: run the same command there, signed into the same GitHub. - Cursor: usage comes from the public profile cursor.com/@username, fetched by this repo's Action. If there is no Cursor data, ask the user whether they use Cursor and for their username, then run `npx @bicrick/tokentracker@latest connect --cursor `. - Data format: FORMAT.md. Stop syncing: `npx @bicrick/tokentracker uninstall`. Delete everything: delete this repo.