Why this is harder than it should be
A typical setup in 2026 is Claude for long reasoning work, Codex or Cursor for editing, and Copilot for completions, plus whatever was interesting this month. Four subscriptions, four dashboards, and four completely different units of measurement.
That is the real problem. Claude gives you a rolling session allowance and a weekly one. Codex gives you a weekly allowance and sometimes hands back reset credits that expire. Cursor counts requests. Copilot counts premium requests separately from ordinary completions. There is no shared unit, so there is no way to compare, and no single place that answers the only question you actually have: can I keep working right now, or am I about to hit a wall?
The three approaches people try
1. Checking each dashboard
This is what most people do, and it fails for a boring reason: it is a snapshot. It tells you what is left, not whether your current pace will get you to the reset. You also have to remember to check, which means you check when you are already worried, which is usually too late.
2. A spreadsheet
Some people log usage manually or export billing data monthly. This works for finance and not for the moment that matters. Nobody updates a spreadsheet at 4pm on a Thursday to find out whether they can start one more refactor.
3. Reading what the tools already write down
This is the approach that actually holds up. The official CLIs and editors already store their login tokens on your machine, and every vendor already exposes the usage endpoint their own app calls. They also write local request logs. A tracker can read those same sources and assemble one view without you doing anything.
What to look for in a tracker. It should read the logins you already have rather than asking for new API keys. It should show reset times, not just remaining percentages. It should project your burn rate so the warning arrives before the wall. And it should never route your credentials through anyone else's server.
The unit that makes tools comparable: money
Sessions, requests and premium requests cannot be compared with each other. Dollars can. Every CLI writes a local log of what it ran, and public model price tables exist, so any of that usage can be priced locally without sending anything anywhere.
This is worth doing even on flat-rate plans, arguably especially on flat-rate plans, because it tells you what your subscription is actually saving you. It is also the only honest way to answer "which of these four am I really getting value from".
Where a tray app fits
The reason a system tray works better than a web dashboard for this is that it removes the remembering. The number lives next to your clock, it refreshes itself every few minutes, and a warning can interrupt you rather than waiting for you to go looking.
Pane is a free AI sub tracker built on exactly this approach. It reads the credential files your official tools already maintain, calls each vendor's own usage endpoint, and puts 21 providers in one popover with reset timers, pace warnings and a shared spend donut. It is MIT licensed, needs no account, and runs on Windows 10 and 11. If you are on macOS, OpenUsage does the same job in the menu bar. Pane is the OpenUsage port for Windows.
Questions people actually type
How can I track all my AI plans at one place?
Use an AI plan tracker that reads the logins you already have, instead of opening each vendor dashboard. Pane does this on Windows: it sits in the tray and shows Claude, Codex, Cursor, Copilot and 17 more in one popover, with reset times and pace warnings.
How can I track all my AI subscriptions in one app?
Install one local app that talks to each vendor's own usage endpoint. Pane is a free AI sub tracker for Windows. OpenUsage does the same job on Mac. Neither one asks you to paste a new API key for tools you already signed into.
What is an AI sub tracker?
An AI sub tracker, or AI plan tracker, is a single view of every AI subscription you pay for: remaining limits, reset clocks, and spend. It exists because Claude, Codex, Cursor and Copilot all count usage in different units, so six dashboards never add up to one answer.
Per-tool guides
How each service counts, and what a tracker can actually show for it:
A short checklist
- Track the reset time for every limit, not just the remaining amount.
- Watch pace, because the useful warning is the one that arrives with a day to spare.
- Price everything in dollars so different tools become comparable.
- Prefer tools that reuse existing logins instead of asking you to mint new API keys.
- Check where your credentials go. They should reach the vendor and nowhere else.