Usage Tracker

A budgeting tool that asks you to trust it with your usage data is a strange thing to build on a site whose whole premise is showing its working. This one doesn't ask that — every request you log here stays in your browser's own local storage, never sent anywhere, which is a real constraint on what the tool can do and a deliberate one.

What it actually logs

Each entry you add is a single Codex or API call — token counts, the model, the service tier — and the tracker keeps a running list in your browser, computed against the same published pricing every other calculator on this site reads from. Nothing about a logged entry leaves the page it's entered on. Close the tab without exporting, and depending on your browser's storage settings, that history may or may not persist to your next visit — this is a lightweight running log for your own reference, not a durable system of record, and it shouldn't be treated as one.

Why local-only, specifically

The alternative — a server-side account that stores your usage history — would let this tool do more: sync across devices, generate reports over time, alert you automatically. It would also mean sending real usage data, which for a lot of people is close to sensitive business information, to a third party that has no operational need to see it in order to do simple arithmetic. A local-only tracker can't do everything a server-backed one could, and that's the trade this site is making deliberately: less capability, in exchange for a genuinely defensible claim that your usage pattern never leaves your machine.

Burn rate, not just a running total

A raw sum of what you've logged answers "what have I spent so far," which is useful but retrospective. The tracker also computes a burn rate from your logged entries — spend per day, extrapolated from your actual logged pattern rather than a number you typed in once and forgot to update — and projects that rate out to a month-end estimate. That projection is only as good as how representative your recent entries are; a burst of unusually heavy usage logged right before you check the projection will skew it upward, the same way a slow week will skew it down. Treat the projection as a moving estimate that gets more trustworthy the more consistently you log, not a fixed forecast.

Where this fits next to the other cost tools

The token-cost estimator and the context-window planner are both forward-looking — priced against a request you're about to make or planning to make. This tracker is the one backward-looking tool on the site: it's for watching what you've actually spent build up in real time, rather than estimating what a hypothetical request would cost. Used together, the estimator tells you what a request should cost before you send it, and the tracker tells you whether your actual logged pattern is tracking that estimate or drifting from it — which is exactly the kind of drift a monthly bill only reveals after the fact.

Clearing your history

Because everything lives in local storage tied to your browser, clearing it is entirely in your control — clear the site's storage the normal way your browser provides, and the log is gone, with nothing to delete on any server because nothing was ever sent to one. That also means switching browsers or devices starts you with an empty log; there's no account to sign into that would carry your history across, which is the direct, visible cost of the local-only design this tool commits to.

Logging manually versus copying from the usage object

The most accurate way to use this tracker is to copy the real figures straight out of the usage block a request actually returns, rather than re-typing an estimate you had in your head before the call went out. That habit also doubles as a running sanity check on every other estimator on this site: if a logged real figure keeps landing meaningfully off what the token-cost estimator predicted for a similar request, that gap is worth investigating on its own, since it usually means the assumed input or output shape doesn't match what a real call actually sends.

Why there's no export-to-spreadsheet feature yet

A tool that logs to local storage and stops there is a smaller promise than one that also hands you a downloadable report, and this site would rather ship the smaller, honestly-scoped version than a half-built export feature with rough edges. If a genuine need for that shows up, it belongs as its own considered addition, not bolted onto a tool whose whole appeal is doing one thing plainly.