spendwatch · service
Find out where your AI coding spend actually goes.
Your team runs Claude Code, Codex, Cursor, Copilot. Each vendor shows its own usage; none shows the others. The result: nobody can answer which project is eating the budget — I found one project quietly consuming 51% of six months of my own spend, and no dashboard could have shown me.
$500fixed price, any team size up to 15
3 daysfrom receiving your data to your report
0 bytesof prompts or code ever leave your machines
How it works
- Free 15-minute preview. One engineer runs
npx spendwatch --json — 100% local, open source, ~700 lines you can read. We look at the output together. If it's boring, we stop here at no cost.
- Your team runs one command each. Every engineer runs the same command and sends me the JSON — token counts, model ids and project paths only. Never logs, never prompts, never code.
- I deliver the report in 3 days. Where the money goes by project, model, engineer and agent; cache health; model-mix; and five specific, ranked fixes with the expected saving of each.
- Optional 30-minute walkthrough with whoever owns the budget.
What you get
- API-equivalent spend — the honest number, whether you're on subscriptions or API billing
- Concentration: which repos and which people account for what share
- Cache-reuse health per team (cache writes cost 12.5× cache reads — most teams have never looked)
- Model mix and where a mid-tier model would cost nothing in quality
- Five fixes, ranked by saving, each with a one-line implementation
- The JSON and the tool, so you can re-run the whole thing yourself next quarter
Booking opens a GitHub issue (no account needed beyond GitHub). Payment by invoice after the preview, before the report — card or bank transfer.