Your First Week With NotchFit: Install and Simulate, Run Real Tasks, Write a Smaller Plan Than You Want, Tune the Alerts, Survive the First Skipped Day, and Decide What Stays

Most people who buy a tool like NotchFit have the same first week: an enthusiastic day one, a day two where the notch keeps opening at the wrong moment, and a day three where they switch it off “for now” and forget it. That arc is not a verdict on the app or on you; it is what happens when a new habit is bolted onto a working day without a plan for the friction. This guide is the plan. It walks through the first seven days with NotchFit, from the install to the point where a Claude Code or Codex run opening the notch feels normal, with one small decision per day and an honest note about what to do when it is not working.

It assumes nothing except a Mac on macOS 14 or later and either Claude Code or OpenAI Codex CLI installed. The reference for every screen is the NotchFit user guide; this post is the pacing, not the manual.

Before you start: what you are actually trying to build

The goal of the first week is not fitness. It is a reflex: agent starts, you look at the notch, you do one short set, the agent finishes, you go back to work. NotchFit is built around that loop. While Claude Code or Codex works, the MacBook notch expands into a panel with a guided, animated exercise on the left and the AI's live progress on the right. Every set starts only when you tap Start set, and if the agent finishes or needs permission while you are mid-set, the alert waits until the set ends. If you measure week one by reps, you will be disappointed; measure it by how many agent runs you noticed and responded to, and it goes well.

Day 1: install, connect, and simulate before you trust it

1. Buy NotchFit on Whop ($20 once, no subscription, no trial, no license key), open your purchase, go to Software and download the signed and notarized DMG. Drag NotchFit to Applications and open it. It appears as a figure icon in the menu bar with no Dock icon, opens at login, and shows the Setup window first.
2. In the Setup tab, click Install / update integrations. That adds the NotchFit Claude Code plugin and the Codex skills and hooks. Restart any Claude Code sessions that were already running. In Codex, run /hooks once and trust the NotchFit hooks.
3. Use the Try it simulate buttons in Setup before you start a real task. They open the notch exactly as a run would, so you can see the panel, tap Start set, and watch the countdown ring without an agent involved. Do this twice. The second time, deliberately leave a set half-finished and trigger a “done” simulation to see the alert wait.

Day one ends there. Do not build a plan yet; the starter plan that exists by default is enough for the first real runs, and the decision about what you actually want to do is better made after you have seen the notch open during work.

Day 2: let it run on real tasks, and do nothing else

Give Claude Code or Codex ordinary work. When the notch opens with the Start workout card and the updates feed, tap Start workout and do whatever the starter plan offers. Use Done, Skip and Pause freely; skipping is not failing, it is information about which exercises fit a desk. Notice two things over the day: how often the notch opened, and how often you were in a state to use it. For most developers the honest answer on day two is “often” and “about half the time”, and both numbers matter for day three.

If the notch opening annoys you on a focused run, that is the moment to learn the Hide workout toggle. It turns the panel into a pure Claude Code and Codex status view, so you keep the done and permission alerts and drop the exercise. Reaching for it is fine; reaching for it every time is the signal that your plan is wrong, not that the idea is.

Day 3: write the plan with your AI, and make it smaller than you want

Now run /notchfit:plan in Claude Code or $notchfit-plan in Codex. Your AI asks about your age, height, weight, fitness level, goal, equipment, days per week, minutes per day and injuries, then writes a weekly plan using only the 239 exercises the notch can animate. The plan is saved to ~/.notchfit/plan.json and NotchFit picks it up instantly. If you would rather stay offline, the Quick plan in the app's Plan tab builds one without any model involved.

The one piece of advice that changes outcomes: answer the “minutes per day” question with half of what you think you can do. A plan you complete on a bad day is the plan that survives week two. You can always ask for a harder one on day ten. Say what your desk allows, too: “no equipment, standing desk, sets under a minute” produces a very different plan from the defaults. We wrote more about this in the AI-written weekly plan.

Day 4: tune the alerts

By day four the exercises are not the problem; the sound is. Open the settings and decide two things. First, whether you want the optional spoken alerts, which use on-device macOS voices and can be muted from the notch; many people keep them on at home and mute them in the office. Second, whether the mid-set wait feels right. It is deliberate: NotchFit never interrupts a rep with a “Claude is done” notification, which is the single feature that stops the workout feeling like a cost to your work. If you find yourself wanting to know the instant a permission prompt arrives, keep sets short rather than fighting the wait; the reasoning is in why mid-set notifications wait.

Day 5: the streak, and the first skipped day

NotchFit keeps a streak, a history heatmap, and throws confetti when you finish a set or a workout. Day five is usually when the streak starts to mean something, and it is also, statistically, when a day gets skipped: a meeting-heavy Friday, a day with no agent runs. Treat the skip as data. If there were no runs, the plan is fine and the day was simply not an agent day. If there were runs and you ignored the notch every time, the sets are too long or the exercises are wrong for your desk. Open the exercise library, look at what you have been skipping, and ask your AI to rewrite the plan without those. Our post on building a movement habit with streaks covers why a flexible streak beats a rigid one.

Day 6: use the quiet moments on purpose

By now the loop runs itself on short tasks. Day six is for the long ones: a refactor, a migration, a test suite the agent is working through for ten minutes. These are the runs where a single set does not fill the gap, and where the notch is most useful, because the progress feed on the right tells you whether the agent is still moving. Do a full workout's worth of sets across the run, with the rest timers between them, and let the done alert bring you back. If you have a standing desk, the movements in standing desk exercises for developers are the ones to put in the plan for these runs.

Day 7: review, and decide what stays

Open the history heatmap. One week is enough to see a pattern: the days that worked, the hours that worked, the exercises that got skipped. Make three decisions. Keep, shorten or replace the plan. Keep or mute the spoken alerts. And decide whether there are sessions, a pairing call, a demo, where you will use Hide workout from the start. Write those three decisions into the plan prompt for next week so the AI builds around them rather than around the defaults.

What to do when it is not working

• The notch never opens. Check that integrations were installed from Setup, that running Claude Code sessions were restarted afterwards, and in Codex that the hooks were trusted via /hooks. The simulate buttons tell you whether the problem is the app or the integration: if simulation works and real runs do not, it is the integration.
• Your Mac has no notch. NotchFit shows a top-center pill instead on a Mac mini, iMac or external display. The flow is identical; see NotchFit without a notch.
• You keep ignoring it. Halve the plan again. The app is a cue; it cannot make a set shorter than you asked for.
• You worry about what it sends. Nothing leaves the Mac. NotchFit runs locally, talks to the agent over localhost only, needs no account, sends no telemetry and never uses the microphone; the hooks run asynchronously so they do not slow the agent. Details are in the privacy post.
• It is simply not for you. That is a legitimate week-one conclusion. The honest trade-off is that NotchFit only helps if your day already contains agent runs; if you use Claude Code twice a week, a plain break-reminder app will cue you more often, and we compared them in NotchFit vs break-reminder apps.

The week in one list

1. Install, connect, simulate twice.
2. Run real tasks; learn Done, Skip, Pause and Hide workout.
3. Write the plan with your AI; ask for half of what you think you can do.
4. Tune spoken alerts; accept the mid-set wait.
5. Watch the streak; treat a skipped day as data.
6. Use a long run for a full workout.
7. Review the heatmap; write next week's plan prompt.

None of this is medical advice, and NotchFit is a workout guide, not a clinician; if a movement hurts, skip it and say so in the next plan. What the first week should leave you with is a reflex that costs nothing to keep: the agent works, you move, the agent tells you when it is done.