Usage trackers for AI coding tools, in the MacBook notch
Notchmeter is one free, open-source Mac app that meters eight AI coding tools beside the MacBook notch — or on any screen edge — with a pace tick, a projection and the reset on every window, and an Advice strip that says what to do about the number. Every figure is the vendor's own or an estimate whose rule is written down, with its source and its date, in docs/accuracy.md.
macOS 15 or later · Apple silicon or Intel · no account, ever
One page per tool
Each page says what Notchmeter shows for that tool, where every figure on it comes from, and what the Advice strip says to do about it.
Claude Code
The 5-hour session window, the weekly window and the per-model weekly limits, plus what your local sessions would have cost at API list prices.
The Claude Code usage trackerClaude Code: when the limit resets
The reset as a countdown or an exact time, whether you will run out before it, and a notification when the window comes back.
When the Claude Code limit resetsClaude Code: the Opus weekly limit
Each per-model weekly limit as its own meter, and one line when a model is nearly gone while another has room.
The Opus weekly limitCodex
The session, weekly or monthly rate-limit windows from the same backend Codex itself asks, and this Mac's sessions priced at OpenAI's rates.
The Codex CLI usage trackerCursor
Included plan usage and on-demand spend for the billing cycle, read the way cursor.com's own dashboard reads it, and the export's own costs.
The Cursor usage trackerGitHub Copilot
The month's AI-credit allowance on a seat GitHub bills by usage, or a legacy seat's premium requests, from the endpoint Copilot's own plugin asks.
The Copilot usage trackerGemini CLI
The per-model quota Google meters for Gemini CLI, from the Code Assist endpoints its own /stats reads.
The Gemini CLI quota trackerAntigravity
The session and weekly quota Antigravity's own panel shows for its model groups, under Antigravity's own identity.
The Antigravity usage trackerKimi Code
The rolling five-hour window, the week and the monthly pools, from the endpoint Kimi Code's own /usage command reads; where Kimi's two kinds of figure disagree about a window, the further-along one is shown.
OpenCode
The OpenCode Go meter, computed on this Mac from OpenCode's own database at the Go page's published rates and limits, its spend on the Cost card and its sessions, with no request to anyone.
How OpenCode is read, in docs/accuracy.mdAll of them, on a Mac
What one meter for every tool looks like, what it reads, what it never does, and what it costs to run.
An AI usage tracker for the MacWhat every meter has in common
The same rules apply to all eight tools, and each is written down with its source and its date in docs/accuracy.md.
Where an even burn would be right now, on every window that declares its length; a window without one shows its fraction and its reset and no tick.
“~67% left at reset” or “Runs out in 2h”. Once there is enough to measure, a window of a day or longer is projected at how far it rose over the last day, and a shorter one gets a run-out range from your own last seven days.
Every window says whose figure it is: the vendor's, Claude Code's status line, a tool's last session file, or an estimate.
Each reading is borrowed from the tool that owns the account, on the login that tool already keeps on this Mac. Nothing is stored and nothing goes to a server of ours.
Which tools report spend
The Cost card carries one row per tool whose spend can be derived from something that tool publishes, and no row at all for one that cannot: not $0, because a zero says "this cost you nothing", and that would be a claim.
- Claude Code and Codex: this Mac's transcripts and session rollouts, priced here at Anthropic's and OpenAI's published rates. An estimate, and labelled as one.
- Cursor: the cost Cursor itself put on each event in its usage export. Nothing is priced here.
- GitHub Copilot: GitHub's own count of the seat's AI credits, a cent each at GitHub's published rate.
- OpenCode: on OpenCode Go, the Go page's published rates applied here to each turn's tokens; otherwise the cost OpenCode itself recorded, or a direct Anthropic or OpenAI model at that vendor's list rate; anything else is counted in tokens with no figure.
- Gemini CLI, Antigravity and Kimi Code: no row. Google and Moonshot meter quota, not money, and publish nothing a rate could be applied to.
The rules, the rates and every known divergence from a bill: Which tools report spend.
Questions
Do I need all eight tools installed?
No. Notchmeter meters whichever tools have a login on this Mac; switch the others off under Settings › Assistants and the strip shows what is left, with the rings giving their room to the ones you keep.
Does it track more than one account per tool?
It reads the login each tool keeps on this Mac, so it follows the account that tool is signed into. Switch accounts in Claude Code, Codex, Cursor, Gemini CLI, Kimi Code or your editor and the notch follows on its next read.
Which tools show a cost?
Claude Code and Codex, priced here from this Mac's transcripts and rollouts at the vendors' published rates and labelled as estimates; Cursor and GitHub Copilot, with the vendor's own figures; OpenCode, from the cost its own database recorded on each turn, or on OpenCode Go the Go page's rates applied here; and Gemini CLI, Antigravity and Kimi Code not at all, because Google and Moonshot meter quota rather than money. A tool that cannot report spend has no row, not a $0.
Does it tell me which tool to use?
When one tool's window is behind pace and another tool on a paid plan still has half of its own left, the Advice strip names it: “Codex has 78% of its weekly left.” It never names a free tier as the room to move to.
Where do the numbers come from?
From the vendor wherever the vendor publishes one, in the same read-only status request that vendor's own app or dashboard makes, and otherwise from a rule written down with its source and its date in docs/accuracy.md, pinned by tests. Notchmeter does not show a number it cannot show its work for.
Install it
macOS 15 Sequoia or later, Apple silicon or Intel. Signed, notarised, and it updates itself. Free forever, MIT licensed.
Notchmeter is a Mac app. There is no Windows or Linux version; if you want a Windows one, the Windows issue is where to say so. How every figure is built, with its source and its date: docs/accuracy.md.