AI Tempomat: too fast or too slow?

Why

When you work with agents on a Claude plan, two questions keep coming up:

• Am I going to hit the limit too early, the way I'm working right now?
• Or could I have started a lot more work in this window?

/usage gives you a percentage, and that answers neither. Whether 62% used is too much depends on how long the window still has to run, and with a few agents going at once it's hard to judge.

AI Tempomat answers both with one number: your pace against the pace that would use the rest of the window exactly at reset. Over 100%, you'll hit the limit before the reset, and the menu bar tells you when. Under 100%, budget will be left over, so there's room for more work if you have it. (Tempomat is German for cruise control.)

What you get

• A gauge in your menu bar. 92% pace, 38% left or 62% used, plus the time you'd run out when you're going too fast: 123% pace · 17:44.
• Four ways to read it. Relative pace like a speedometer, or Total like a fuel gauge. Averaged over the whole window, or over the last 30 minutes so it reacts to what you're doing now.
• Which agent burns what. The expanded view draws every Claude Code session as a ribbon of tokens per minute, lists what's running now with its open tasks and subagents, and charts where the limit is heading.
• Every window, kept. The History page keeps each 5-hour window: how much you used, what you left unused at reset, when you hit the limit and how long you waited. Spread one out to find its peaks and what burnt them, down to the subagent and the prompt.
• Local, and free to run. Limits come from claude -p "/usage", which never calls the model and costs no quota. Tokens come from Claude Code's transcripts in ~/.claude. Your usage and tokens stay on your Mac: the only thing the app sends is a check for updates at launch and once a day, with its version numbers, the macOS version, its branch and four yes/no flags, and no ID. You can turn it off.

Ethical considerations

Don't start agents just so the budget isn't "wasted".

The dial shows when there's room left in the window. That doesn't mean the room has to be filled. Unused budget isn't waste: a window that resets with 20% left used no extra energy at all.

Tokens that get spent do. Every request runs on GPUs in a data center that draw electricity and need cooling, and depending on where and when that power is generated, it adds to carbon emissions and water use. Spent on real work, that can be worth it. Spent to fill a quota, it isn't.

So use the room when there's work waiting that's worth doing, and let the window reset when there isn't. Treat 100% as a ceiling to plan against, not a score to hit.

Install

Paste this into Terminal:

curl -fsSL https://raw.githubusercontent.com/Taremeh/tempomat-ai/main/scripts/install.sh | sh

Then click the gauge in the menu bar to open the panel, and click Sign in. Your browser opens claude.com: approve there, and the panel shows your limits.

You need macOS 14 or later; building it takes Apple's Swift 6 tools, which need macOS 14.5 or later. The script installs Apple's command line tools if they're missing (macOS asks first), clones this repository into ~/tempomat-ai, builds the app, puts it in ~/Applications and starts it. If Claude Code isn't installed, the panel offers Install Claude Code, which runs Anthropic's installer in a Terminal window. The app needs Claude Code 2.1.29x or later.

Keep the clone: the app's Update button pulls and builds there (see Updating), and running the script again does the same. To clone somewhere else, set TEMPOMAT_DIR. A folder outside Documents and Desktop spares you a macOS prompt on updates:

curl -fsSL https://raw.githubusercontent.com/Taremeh/tempomat-ai/main/scripts/install.sh | TEMPOMAT_DIR=~/src/tempomat-ai sh

The script never deletes or overwrites a folder of yours: if that folder is already there and isn't a clone of this repository, it stops and says so. Like ./scripts/build-app.sh --install, it does replace ~/Applications/AI Tempomat.app. To read it first, see scripts/install.sh.

Signing in

The app reads usage by running claude -p "/usage". This is a local command, so it never calls the model and costs no quota. It needs Claude Code signed in to your Claude plan, and the desktop app's login isn't shared with the terminal CLI. Until the CLI is signed in, the panel says so and offers Sign in.

Sign in runs claude auth login --claudeai for you, which opens claude.com in your browser. Approve there, and the panel shows your limits a moment later. If the browser didn't open, the panel links to the sign-in page. If the CLI is signed in with a Console account (API billing), the panel offers Sign in too. Signing in replaces that Console login with your Claude plan's login, for claude in Terminal as well.

You can also sign in from Terminal instead (if claude isn't found there, the panel shows the full path of the copy it uses):

claude auth login

Install by hand

If you'd rather run each step yourself, install the Xcode command line tools (xcode-select --install), then clone, build, install and launch the app:

git clone https://github.com/Taremeh/tempomat-ai.git
cd tempomat-ai
./scripts/build-app.sh --install

Then sign in from the panel, or with claude auth login, as above.

In the clone (~/tempomat-ai if the script made it), check that the CLI side works without the UI:

.build/release/AITempomat --print-usage

Run the tests:

./scripts/test.sh

여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗